XScale and the Pocket PC – what’s going on?

[Jason Dunn]

With all the anticipation over the Intel XScale processor running at 400 MHz, expectations were high that any Pocket PC would double in speed. You double the MHz from 206 to 400 and the Pocket PC should get twice as fast, right? Not exactly. There are many issues that relate to overall device speed, not the least of which is software. Ed Suwanjindar from the Microsoft Mobile Devices group responded to my questions via email on this issue. I'll post my own thoughts on this issue under a separate entry. On to the Q&amp;A! <!><br /><br /><i><b>THOUGHTS: Early reports based on those who own the Toshiba e740 Pocket PC 2002 device are telling us that XScale at 400 MHz performs slower than a StrongARM at 206 MHz on some tasks. This came as a surprise to many people.</b></i><br /><br /><b>SUWANJINDAR</b>: “We are aware that PXA250 (XScale)-based devices are not demonstrating the huge performance gains that were anticipated. That said, Pocket PCs continue to offer the best performance and the richest functionality vs. other handhelds on the market today.”<br /><br /><b><i>THOUGHTS: I’ve seen a few articles on line saying it's Microsoft's fault for not having an optimized OS in place for the XScale launch. What is Microsoft’s response to this?</i></b><br /><br /><b>SUWANJINDAR</b>: “Our software remains the same. This is the same Pocket PC 2002 software that performs fabulously across other ARM processors (StrongARM 1110, OMAP710, etc). We made a hard decision several years ago to move away from supporting several processor architectures and target a single core. This was a difficult decision that we think ultimately benefited our OEMs, developers and customers by unifying our platform around single processor architecture -- ARM V4. The PXA250 utilizes the ARM V5 instruction set with backwards compatibility for ARM V4. When we completed the Pocket PC 2002 software in June 2002, we optimized for the most broadly compatible processor core available at the time (ARM V4), which it still remains today. Choosing to support one processor core ensures we don’t fragment our platform for developers and cause extra work for our ISVs to optimize their applications each time a new processor technology is released.<br /><br />By staying with ARM V4 architecture we assure longer life spans for our customers existing hardware – for instance if we were to move to an ARM V5 architecture we would have to obsolete the all SA1110 iPAQ devices. Protecting the investments of our developers and customers is very important to us. To that end we’ve worked to make our devices upgradeable. Moving to ARM V5 would break upgrade compatibility. We’re not prepared to strand an installed base of over 2 million iPAQ users.”<br /><br /><i><b>THOUGHTS: Some industry analysts have said that Microsoft doesn't have any fix in place because Intel couldn't get the chips out in time.</b></i><br /><br /><b>SUWANJINDAR</b>: “We have implemented and released specific software changes that our hardware partners are implementing without breaking compatibility for our OEMs and users. While we believe there may be incremental gains that could be had via small optimizations we are not convinced there are across the board improvements that would amount to any kind of dramatic system wide speed up. We have to develop software based on the processor architecture that offers the broadest compatibility for developers and when we shipped Pocket PC 2002 as it still is today, that was ARM V4.”<br /><br /><i><b>THOUGHTS: Some of those same analysts have said it will be 2004 until there's an OS that can use the XScale CPU properly. Is that an accurate estimate?</b></i><br /><br /><b>SUWANJINDAR</b>: “It’s too early to talk about the next version of our software. That said, we’re committed to delivering best-in-class functionality and performance while providing a foundation that enables our developer community to continue to innovate and build successful businesses on our platform.<br /><br />Microsoft considers mobile devices a strategic business. We are committed to working closely with Intel and other silicon vendors on delivering future versions of our Pocket PC and Smartphone devices. We have released specific software modifications to our OEMs that in total are all of the optimizations we believe are possible to maximize PXA250 performance (without causing incompatibilities for our OEMs and developers).”<br /><br /><b><i>THOUGHTS: This isn’t a good story for the Pocket PC and as more XScale devices hit the market, the issue will get more obvious and ultimately become more serious.</i></b><br /><br /><b>SUWANJINDAR</b>: “Agreed, this isn’t a good story. Very simply, we think this is one of those times when the technical reality didn’t measure up to market expectations. That said for people who use these products, this isn’t a big deal. I’ve used both of the new XScale products that are out there (new Toshiba and iPAQ). They offer the same type of performance that I’ve come to expect on a Pocket PC. I think the market expectation of what performance on a 400 MHz processor vs. 206 MHz processor has been unreasonable. In the mobile device space, we don’t think that MHz is what ultimately matters to customers. What matters the most in this market is whether customers can do what they want to do with devices quickly and easily. With the richest set of software applications built into any PDA on the market, and the strong momentum that Pocket PC has with developers writing for our platform, we think that customers will be able to do the things they want to do with the performance they expect on devices using PXA250 processors.”

[PJE]

I read somewhere that the new instructions in the ARM v5 core would cause and exception within the processor which would act like an interrupt. Software could then decode the specific instruction (although slower than a v5 core).

This was originally installed (I think) to allow devices with and without a floating point unit, where the missing hardware could be emulated in software with a subsequent performance hit.

I think it's definately in Microsoft's interest to get the PocketPC running full tilt on XScale, otherwise Palm V6 may come along and eat it for breakfast.

Regards,

PJE

[Ed Hansberry]

I think it's definately in Microsoft's interest to get the PocketPC running full tilt on XScale, otherwise Palm V6 may come along and eat it for breakfast.
I don't think Palm is V5 ready anymore than PPC is. Both are optimized for V4 ARM code. All of Palm's partners are producing ARM V4 chips anyway. They just announced they will run on the X-Scale, but the didn't do anything for that. V5 runs V4, it just runs slow, negating much of the speed gains. :(

[mookie123]

I don't think that's the kind of answer that I want to hear.

I basically says. Sorry we won't do V5. Optimizing V5 globally won't get us that much faster anyway.

This is the kind of thing PALM Inc. would spew! gargh!

Conclusion: Buy el cheapo SA PPC, they give more bang for the buck.

[Foo Fighter]

Sounds like we're screwed. Microsoft has no intention of "optimizing" the OS for xScale. :oops:

[ChrisD]

While this story explains why Microsoft was not ready with ARM v5 support, it does not explain the peformance issues. I believe that articles like these dance around the real issues that are holding back performance.

Overall, Intel's decision to continue using a 100 mhz bus for ram and peripherals with a 400 mhz cpu was a step backwards from the 103 mhz bus with a 206 mhz cpu. So until Intel adds support for faster memory access these chips will not perform at a full 400 mhz all the time.

See "Are Pocket PC's Starving?" for a more technical discussion about these issues at http://www.cewindows.net/faqs/pocketpcstaving.htm

[donkthemagicllama]

This blows.
Clearly PIM apps etc. won't benefit from the XScale, but can multimedia apps that would stand to gain from increased performance be written to take advantage of XScale, or will the OS prohibit that?

In other words, we know that PPC2002 doesn't and won't take advantage of XScale, but does that mean 3rd party apps can't either? Or can they be written that way at the expense of backwards compatibility?

Anyone?

[jpaq]

Translation:

We forgot to keep the url www.pocketpc.com registered in the middle of a court case involving the product. You think we'd have our sh*@ together enough to make a hardware agnostic OS? Then again, we can't/won't make our own word processing program for PPC2002 work as well as competitors on Palm. Tables? You don't need no stinkin' tables! If that doesn't tell you something.....

That would make too d%$ much sense.

Yep. PPC2002 runs on Xscale. Not worth a crap and in some cases worse than old technology, but it runs........kinda.



Microsoft and the good ole boy network of PDA manufacturers have, effectively, built the largest beta test group ever.

Don't get me wrong. I wouldn't go for the alternative (Palm), but it had to be said.

Thinking..........

Those new Sony clamshell PDA's are nice.

(Smacks himself) SNAP OUT OF IT!

Couldn't MS just insert processor sniffing OS code and let US and the developers choose what we want. Maybe even offer V4 compatibility mode in a V% optimized version of the OS (See windows XP).

I'm babbling now, but it makes sense.........I think.

:?

[Speed Racer]

That sounds like a lame excuse for the poor performance. It sounds like Palm's old line where they said that nobody would want to have color or multimedia on a PDA. In reality it was just a weak defense for their PDAs that didn't provide those features.

THOUGHTS: This isn’t a good story for the Pocket PC and as more XScale devices hit the market, the issue will get more obvious and ultimately become more serious.

SUWANJINDAR: “Agreed, this isn’t a good story. Very simply, we think this is one of those times when the technical reality didn’t measure up to market expectations. That said for people who use these products, this isn’t a big deal. I’ve used both of the new XScale products that are out there (new Toshiba and iPAQ). They offer the same type of performance that I’ve come to expect on a Pocket PC. I think the market expectation of what performance on a 400 MHz processor vs. 206 MHz processor has been unreasonable. In the mobile device space, we don’t think that MHz is what ultimately matters to customers. What matters the most in this market is whether customers can do what they want to do with devices quickly and easily. With the richest set of software applications built into any PDA on the market, and the strong momentum that Pocket PC has with developers writing for our platform, we think that customers will be able to do the things they want to do with the performance they expect on devices using PXA250 processors.”

I don't know about anyone else but I can see a noticeable improvement by overclocking my iPaq to 236 MHZ. Everything seems to run a little smoother and the frequency of slow screen redrawing and program pauses is noticeable reduced. I for one was expecting a marked improvement over my existing iPaq because the improvements mentioned above were realized with only a 30 MHZ increase. Shouldn't we expect a noticeable improvement if the clock speed doubles? Likewise I don't think that PocketPC has reached its full potential and Microsoft should not be intentionally plateauing its performance.

[Charles Pickrell]

Intel was the exclusive Pocket PC processor manufacturer. The fact they didn't take the rather large V4 ARM codebase into consideration during processor designs and testing shows they are totally clueless. If the Pocket PC 2002 apps, Windows CE.NET apps, Palm V5 apps, and smartphone apps all run more slowly on this thing why would anyone want to buy it at all? Give me the old ARM 206 processor anyday.

Charles

[Foo Fighter]

As I've said before, I don't believe Microsoft really takes the PDA market seriosly. This xScale fiasco proves it. Microsoft's mobile division is a revolving door for top-level executives. Even Ballmer himself headed it at one point. They go through more executives than Generals in the Union Army.

If Palm manages to deliver smoother multimedia performance with OS5, what is Microsoft prepared to do in order to stay competitive with its slower performing ARM4 crippled PocketPC platform? My guess: Not a damn thing. They are more concerned with the enterprise market, and integrating PPC into the .NET buracracy to care what a bunch of whiny geeks complaining about slow video quality think. :roll:

SUWANJINDAR offered nothing more than petty excuses to explain away the ugly ills of the platform.

[grogma]

OK, let's look at this the other way around. Why the hell are the PocketPC manufacturers going nuts churning out Xscale based PocketPCs when there's no appreciable benefit? Is this all just marketing hype designed to make me unhappy with my wimpy old 206MHz SA based machine? Its not like the PocketPC manufacturers didn't know that the old SA code would run poorly on the new architecture. If Microsoft isn't going to have an Xscale optimized PPC OS on the market until the vast majority of the installed base is Xscale then the only appreciable benefit of buying an Xscale based PPC NOW is to build up the user base of Xscale machines. Take one for the team! Buy an Xscale so Microsoft will port PPC to ARM V5! But not for your unit because it will be long discontinued by then! We've got a loooooong wait ahead of us. Argh.

[Foo Fighter]

In other news: HP is going to unveil new iPaqs on Monday, starting at $650.

I wonder how many PPC users are going to spend this kind of money to "trade up" to a device that is slower than the one it replaces.

This is going to blow up in Microsoft's face.

[jim s]

We are aware that PXA250 (XScale)-based devices are not demonstrating the huge performance gains that were anticipated. That said, Pocket PCs continue to offer the best performance and the richest functionality vs. other handhelds on the market today.

Translation:
We know your new car only goes 40mph instead of the 65mph you old car did, but it beats a bicycle, doesn't it?

[Sslixtis]

Charles Pickrell Posted: Thu Jun 20, 2002 2:39 pm Post subject:

--------------------------------------------------------------------------------

Intel was the exclusive Pocket PC processor manufacturer. The fact they didn't take the rather large V4 ARM codebase into consideration during processor designs and testing shows they are totally clueless. If the Pocket PC 2002 apps, Windows CE.NET apps, Palm V5 apps, and smartphone apps all run more slowly on this thing why would anyone want to buy it at all? Give me the old ARM 206 processor anyday.



Yeah, Da** Intel anyway! Why would they want to go and produce the next generation of hardware?!? Lets wait on MS to tell the Hardware manufacturers when to make new chips!!! What a load of CRAP! Hardware always drives software, Intel could have stopped with the 8088 processor "all the software was optimized for it" but aren't you glad they didn't? I am. Why do people insist on blaming the hardware manufacturers for lazy software companies? When MS and the other software companies start to really push Arm V5 software everyone will say how wonderful they are and conveniently forget that it was the HARDWARE that drove all those wonderful improvments. Didn't this company called Palm or something have this issue with making improvements to its handhelds until a new kid on the block came along and started innovating the handheld market? Am I going to have to switch to a Linux PDA to get the performance I want? Bottom line-INNOVATE OR DIE!

[jmulder]

I think it's interesting that we're all bad mouthing Microsoft for not optimizing their code (probably because the interview was with a rep from Microsoft), but what about Intel?

Why should Microsoft have to change? Shouldn't Intel have taken the time to ensure that V4 code could take advantage of the increased clock speed? Isn't it just as irresponsible of Intel to assume Microsoft would foot the bill on this?

I'm not saying I subscribe to this point of view...I just want to point out that it exists. Feel free to blow holes in my logic. (Just don't give me any bull like 'it's harder to change firmware')

-Jim

[denivan]

I may be getting this wrong, but isn't this more intel's fault, than it is Microsoft's fault? If they made an optimized OS for the xscale cpu, wouldn't that also mean we would go back to the time that there are different versions for every program? Now, finally, the different version (MIPS, SH3...) are starting to phase out, so most new programs are arm only. But if an optimized pocketpc os would come out for the xscale, wouldn't that mean that all programs should be rewritten for this optimized os ? i might be wrong, but it's just a thought ;) Anywayz, I'm not Mhz hungry when it comes to my ipaq, so i would just be happy if the xscale would save alot of battery power, actually to me, that's what matters most...

[alex_kac]

My one addition to this is that device mfgs did not know until just a couple months ago how badly the XScale performs compared to its expectation. The hardware, software, etc... was all ready for it. So you can say the OEMs aren't happy about this either since it puts them into a pickle. I don't think MS knew either until a couple months ago either.

There was NO production quality 400Mhz XScale processor available until a couple months ago to test on.

It is my opinion (and a strong one) that Intel really screwed the ARM world over....

[mja8105]

In the mobile device space, we don’t think that MHz is what ultimately matters to customers. What matters the most in this market is whether customers can do what they want to do with devices quickly and easily. With the richest set of software applications built into any PDA on the market, and the strong momentum that Pocket PC has with developers writing for our platform, we think that customers will be able to do the things they want to do with the performance they expect on devices using PXA250 processors

Remind me again, aren't you going to need higher MHz to run richer apps?

[Robotbeat]

Ummm...

Microsoft knows that optimizing the OS for XScale doesn't matter more than about a 1% increase in multimedia intensive apps. Compatibility is more important than a 1% increase in performance, don't you think?

All Microsoft has to do is release the right SDKs and development tools. They have optimization for XScale in their CE.NET EVC4, why not for PPC 2K2? They probably are working on throwing together a PPC 2k2 SDK for EVC4 right now. Not too hard to do, when you actually think about it. That said, it probably matters more about the other hardware crap like how the iPAQ 3800 was all messed up in the way it did GAPI. I'll just wait to see which OEM fixes their display drivers first and then I'll get an XScale PPC.

Personally, I want to see how fast an e740 does a CPU-intensive but non-GAPI app like PocketYASRT (http://www.yasrt.org/) (a simple ray-tracer) or something. I'll find out...

[Sslixtis]

Why blame Intel, it is the same thing as every other new generation of chip. The hardware comes out, it isn't as great as everyone expects but it IS backwards compatible, BACKWARDS being the operative word. The software companies write optimized software and everything is right with the world. Perfect example is the P3/P4. Same thing happened with the chip set except MS stepped up and Immediately had a OS optimized for it. I agree with Foo Fighter MS doesn't take the PDA market as seriously as they do the Desktop market. More profit in the desktop market by a long shot but that market is becoming overly saturated and the PDA market is realitive babe. Anyway, all of you guys whining and ragging on Intel for pushing the hardware envelope could always switch to Palm if it really bothers you that much :wink: Personally I will stick with Intel and the X-Scale. By the way, the new Samsung processors for the Handheld market that are clocking 500-750Mhz are ARM V5 also, so I wonder if they are going to decide to scrap their chips and go back to the ARM V4...NAH, and what about the 1Ghz X-Scale in the Q1-Q2 of 2003 why not scrap it and stick with the 206 StrongARM, it would be too much trouble to Optimize the software. Who wants true speak recognition anyway and NOBODY wants MultiMedia on a Handheld didn't Palm tell us so a long time ago?

[Sslixtis]

Yeah, and the we aren't supporting it because no one has it line smacks of concern for the consumer doesn't it?!? Good thing they didn't say that when the 286, 386, 486, Pentium, Pentium II, Pentium III and Pentium 4 all came out! It must be different somehow. Yeah, I see now for those chip sets they didn't CARE about the customer NOW they do :wink: That must be it! Whew, I really wish they would go back to not CARING, it was alot more productive.

[LarDude]

I heard some radical theories that Ed Suwanjindar and Micheal Mace are actually the same person. After reading Jason's interview, I am now starting to believe it. Can it be long before Oliver Stone makes a movie about this? :wink:

[pt]

Wow, I'm really impressed with the latest here at pocketpcthoughts.com

1. The CF / SD article and interview was great.
2. And this interview was also great. Way to go!

Microsoft didn't need to do this interview, they're speaking to us in our community, in a human voice, the best they can. I think this is a great trend and look forward to it continuing here.

Well done.

Now, on to the topic. I'm glad all my Pocket PC 2002 software will run on my Xscale Pocket PC, and get some better battery life. And when there are optimizations down the road, I can move towards those.

Maybe I'm used to this since I have a 800Mhz iMac. While that number sounds slow compared to my 2Ghz Pentium, the iMac is better at certain tasks.

[LarDude]

:( I am so deflated with all the recent news about Xscale performance. For so long, so many of us have been so excited about the imminent release of Xscale PDAs...and now this...talk about Pockus Interuptus.

I was ready to fork over the ~$1000 Cdn for one of these babies, now I wonder how long I should put off buying. By the time the Xscale-optimized OS fix comes along, how much will the hardware requirements have changed? Will 32-48MB ROM still be enough? Will something as trivial as the built-in Microphone still be good enough (will a voice-input interface with a noise-reduction uni or omni-directional mic
be the recommended minumum by then)? Would the NR70 have forced M$ to go to a different screen rez?

[DJR]

I agree that it was great that a Microsoft rep has made a public response to this issue in this forum; a little too much "marketing-speak" in there, but informative nonetheless.

I wonder if the decision to lock the PocketPC platform to a single processor has turned out to be such a wise one. PalmOS has a "PalmOS ready" program for running across a spectrum of ARM-architecture processors (TI's OMAP, Motorola MX1 and Intel XScale; plus I think that MediaQ is now certified). Of course, there won't be any PalmOS ARM devices shipping for a few weeks (?) but it's going to be very interesting to see how the hardware compares... I recently saw a demo of a movie player on a Sony Clie - 480x320 full color at 15fps and it was pretty impressive for a 66MHz Dragonball processor...

PS I've got the Toshiba support guys hunting for the Toshiba SDIO bluetooth drivers bt, and will forward them if they have any luck!

[fmcpherson]

It should be clear that Microsoft should have never shipped Pocket PC 2002. They should have held off releasing a new Pocket PC until Xcale was coming to market and then timed the new software with the new chip.

However, the downside side is that all iPAQ 3600 owners probably would not have been able to upgrade to Pocket CP 2003.

[mookie123]

Questions:

1.Chicken and egg problem. Do you think customers will be compelled buying new models if microsoft doesn't feel like optimising for V5, and waiting for critical mass. Won't developer be also waiting for the critical mass to start writing spiffy apps? When does Microsoft feel is the right time to implement optimisation?

2. Isn't there a way to maintain previous copatibility while adding new features for the new machine? Or are we really hang up on those 2 millions iPAQ out there? That all new softwares MUST run on all those iPAQs.

3. Are ARM plc and Intel Inc bunch of complete moron wasting silicon for some insignificant features?

I think somebody should voice it to Microsoft that they are making a mistake here, that consumer will NOT accept this sort of bogus solution. (you buy it first, than maybe........JUST maybe we might optimize it...)

How would they like hearing:......I won't BUY those spiffy gimmick until you fix it.



PS. don't you hate it when somebody just pee on your shoe and call it 'rain?

[DJR]

Why the hell are the PocketPC manufacturers going nuts churning out Xscale based PocketPCs when there's no appreciable benefit? Is this all just marketing hype designed to make me unhappy with my wimpy old 206MHz SA based machine?

Apparently :?

[Ed Hansberry]

Remind me again, aren't you going to need higher MHz to run richer apps?
I think that would depend on the app. Games? Sure. Excel? Well, I have Excel XP on both a Pentium II-366 and a P-4-1.9GHz. I am a heavy Excel user and honestly cannot tell much difference unless I do something like run a big pivot table or load something with several thousand formulas.

I think ChrisD hit the nail on the head in many respects. The bus speed is the bottleneck. A 1GHz processor isn't going to do much until the bus is fatter and faster. :(

[Ed Hansberry]

It should be clear that Microsoft should have never shipped Pocket PC 2002. They should have held off releasing a new Pocket PC until Xcale was coming to market and then timed the new software with the new chip.

However, the downside side is that all iPAQ 3600 owners probably would not have been able to upgrade to Pocket CP 2003.
Oh yeah. I agree completely. :roll:
I would have much rather forgone VPN, terminal server, the spell checker, inbox sub folders and several other things while MS waited for Intel to actually finish the processor and then start working on it.

Why is it Intel does a poor job with ARM V4 code processing on their V5 chip MS's or any software makers fault? :?:

[Jonathan1]

The PXA250 utilizes the ARM V5 instruction set with backwards compatibility for ARM V4. When we completed the Pocket PC 2002 software in June 2002, we optimized for the most broadly compatible processor core available at the time (ARM V4), which it still remains today. Choosing to support one processor core ensures we don’t fragment our platform for developers and cause extra work for our ISVs to optimize their applications each time a new processor technology is released.


BS total BS :evil:

I brought this up over at PDA BUZZ........

Been going over intel's documentation for the PXA250

On document Intel® PXA250 / PXA210 Applications Processor Design Guide that is 278523-001.pdf its revision history is:

Revision History
Date Revision Description
Nov 2000 0.1 Initial Release: RS-Intel® PXA250 Platform Design Guide
Nov 2000 0.2 Second draft
Jan 2001 0.3 Corrected name of FFRTS in Table 1-4. Reorganized Table 1-4 and Table 1-5 for readability.
May 2001 0.6 Added reference to PXA210 and performed editorial clean-up.
February 2002 1.0 Public Release



As I said over there is "SEEMS" as if MS knew about the X-Scale a long time ago. At minimum Early 2001.

This is pure MS FUD. These people have been caught with their pants down and they have to give ANY excuse other then admit ya we screwed ourselves. (To be fair Palm would/has done the same.) It isn't the fact that MS DIDN'T working on this problem a long time ago that is steaming me, but more that the fact that they KNEW about this problem for more then a year. No one here can tell me that MS doesn't have resources at Intel to know that they were migrating to V5 of the instruction set.

This FUD about supporting older devices is BS and Microsoft knows it. V5 of the ARM core supports v4 so there is total backwards compatibility. So older software would work on newer devices and lets face facts. The next version of PPC is going to be CE.NET based. I doubt there is a snowballs chance in heck of fitting it in the 36xx’s pathetic 16MB of ROM or even in the 37xx –38xx 32MB of ROM either. So there is no reason why CE.NET version of PPC shouldn’t just go for the gold and be optimized for X-Scale.

Finally I have to question just how different the cores between v4 and 5 are. How difficult are they to migrate from one to another (i.e. Compile to run.) I don’t know squat about programming but if the tools are there then there is no reason not to give the next version of PPC a kick in the pants with optimized code. This REEKS of the same crap coming out of the Palm camp. OS5 runs on ARM but it isn’t any great performance boost because its not ARM optimized.

Again I’m weak from a programming standpoint. If the OS doesn’t support X-Scale optimization can the programmer skirt around this and still use the power of the X-Scale or do development SDK’s need to be implemented before this happens.

The BS at MS is so high you can smell it across the net. :twisted: And guess who’s gonna get dumped on in the end *opens an umbrella*

[fireflyrsmr]

[Microsoft didn't need to do this interview, they're speaking to us in our community, in a human voice, the best they can. I think this is a great trend and look forward to it continuing here.]

pt has it right. it's remarkable and refreshing that they are talking about this. I have to say that a lot of responses seem amazingly narrow in their perspective. if they could have, they would have - brought out a better product. now that they are disappointed and telling us so we need to wait for them to get it better which they will do as soon as they can see sense in it. i'm glad that didn't trash me with the other current ipaq owners. maybe some of you have some big needs (or more games :twisted: ) but i thought he made sense that all the app's that make my corporate life better, run plenty fast. I don't want them to get lazy and you guys with the noose and bonfire are probably a good thing in that respect but it was getting a little silly. someone who needs a pda who hasn't jumped in yet will be happy to get the 400 mz machine.

while i'm at it jason - the flash interview was wonderful too.

[Jason Dunn]

Front page rant coming up tomorrow. Intel had better run. :twisted:

[entropy1980]

Remind me again, aren't you going to need higher MHz to run richer apps?
I think that would depend on the app. Games? Sure. Excel? Well, I have Excel XP on both a Pentium II-366 and a P-4-1.9GHz. I am a heavy Excel user and honestly cannot tell much difference unless I do something like run a big pivot table or load something with several thousand formulas.

I think ChrisD hit the nail on the head in many respects. The bus speed is the bottleneck. A 1GHz processor isn't going to do much until the bus is fatter and faster. :(


Balogna! While I agree faster bus speed would help think about this Pentium 3's until recently were out perfroming Pentium 4's at nearly double the clock speed and with a much faster bus.... why? Software , the software does not take advantage of the 20 stage pipeline the P4's have. I don't blame Intel fully nor do I blame Microsoft fully, I blame them both for lack of communication that is the problem everyone wants to pass the buck. The bottom line is even with a faster bus I doubt we'd see much of any gain in performance.

[Brad Adrian]

It should be clear that Microsoft should have never shipped Pocket PC 2002.

wow. That seems pretty harsh. I have been using the new 2002 features to DEATH since I got the upgrade. Sure, there have been problems, but to say that nothing should be rolled out unless it's perfect or has all the desired functionality seems pretty unrealistic.

[Timothy Rapson]

SO the two largest computer companies in the world by far, Intel and Microsoft, can conspire to set up a PDA world where HP sells (tries to sell!) new Ipaqs for $800, but they can't deliver a working CPU upgrade with 19 months lead time! Pathetic, just pathetic.

SUWANJINDAR sounds like some politician. He has an answer to every question. It all sounds so good while he is talking. He cares about PPC owners, ISVs, the future, partners, consumers, everyone.

But the bottom line is that PPC is now hung out to dry.

Intel and MS and all the PPC manufacturers have known about X-Scale for a LONG time. I can't remember how long ago Intel and MS were all watermelon smiles as they bragged about the promised land they were looking over the mountain at. LOOK! they said. We can see it so clear. Look at this demo. See how fast it will run on the new processors! Pay no attention to the fact that it is a staged demo of "expected performance inprovements." How could they take this long to deliver this little?


The new processors are here, but worthless. The production has probably already stopped on the older ones. They might have been cheaper if they didn't have to pay for the redesign costs whether it is useful or not. But, the StrongArm 206MZ is going to be gone.

In a year or more we may (or may not----- remember those great built-in applications means that no one else it going to ship a nice word processor or video player that might be optimized for the X-Scale) get new applicatioins or an OS that uses the 400-1000 MZ. By then Sony will have 600 MZ TI ARMS running Palm OS6 with built-in wireless, telphoney, and video cameras.

And it could have been so nice to have a PPC with 600MZ running a full VGA screen, camera, long battery life. Bummer.

[mormondad]

My optimism/denial is blown. Argh. I've been waiting two years for this PocketPC and I HOPED that the Xscale issue would be fixed quickly. Little did I know that Microsoft doesn't give a flying #%$@ about the problem and they think we should be happy that we just spent $600 on something that is no better than the stuff that costs half as much just because some %#$@head engineer thinks it works just fine for what he thinks it should work for.

[Dan East]

This is not an OS issue. The performance problems we see with games like Pocket Quake can only be a result of hardware issues. It is interesting that we haven't seen the first benchmark related to XScale. Every time a new x86 processor is released we see tons of benchmarks illustrating each drop of computing power it exerts beyond that of its predecessors. Why isn't XScale being sold in the same manner, instead of quoting only its clock speed? I think everyone from the OEMs to Microsoft to Intel has known all along that we would see virtually the same performance with the next generation Pocket PCs as we did when the first iPaq 36xx was released years ago.

Dan East

[Ed Hansberry]

BS total BS :evil:

I brought this up over at PDA BUZZ........

Been going over intel's documentation for the PXA250

On document Intel® PXA250 / PXA210 Applications Processor Design Guide that is 278523-001.pdf its revision history is:

Revision History
Date Revision Description
Nov 2000 0.1 Initial Release: RS-Intel® PXA250 Platform Design Guide
Nov 2000 0.2 Second draft
Jan 2001 0.3 Corrected name of FFRTS in Table 1-4. Reorganized Table 1-4 and Table 1-5 for readability.
May 2001 0.6 Added reference to PXA210 and performed editorial clean-up.
February 2002 1.0 Public Release



As I said over there is "SEEMS" as if MS knew about the X-Scale a long time ago. At minimum Early 2001.
It seems as if Intel is a bit surprised at how poorely it is performing with ARM4 code. MS had books and papers in 2000. big whoop. That isn't the same thing as a working product. of course MS knew about X-scale, but if they didn't have one until late 2001/early 2002, this is their fault how?

[ppcsurfr]

HAs anyone tested the Xscle devices for WMV files? I know they've tested it for MPEG files which don't really require much CPU power, but would it be possible that the new Xscale processors be more capable of handling WMV codecs?

How about multitasking? Maybe the Xscale devices hold up better even during multitasking?

[mormondad]

HAs anyone tested the Xscle devices for WMV files? I know they've tested it for MPEG files which don't really require much CPU power, but would it be possible that the new Xscale processors be more capable of handling WMV codecs?

How about multitasking? Maybe the Xscale devices hold up better even during multitasking?

WMV files played with Windows Media Player show the worst performance of all. MPEG works OK in PocketTV (though we did expect improvement) and some files work passably in DivXPlayer.

[/dev/niall]

Microsoft didn't need to do this interview, they're speaking to us in our community, in a human voice, the best they can. I think this is a great trend and look forward to it continuing here.


Yeah, but... he didn't really say anything. I mean, replace a few nouns and he could be talking about anything. Seems like standard marketing drivel to me.

[crispeto]

I gotta be honest. I bought a new Maestro about three weeks ago and now I'm glad I did. I debated between buying one now or waiting for Xscale. The Maestro $299 price did it for me. I can't imagine paying $800 now for possibly worse performance. Looks like I dodged the Intel/MS bullet there.

[Nikhil]

Sadly enough, my iPAQ 3835 outperforms my Toshiba e740 in multitasking (including I.E. browsing...the e740 slows down significantly, the iPAQ does, but not remotely to the same extent). Same with basic program multi-tasking...The iPAQ outperforms in multimedia..The only speed increases I've found are in load times for Tennis Addict, Lemonade Inc., Adobe Acrobat Reader, and Microsoft Reader. Once the apps are loaded, however, both units perform identically.

[pt]

Yeah, but... he didn't really say anything. I mean, replace a few nouns and he could be talking about anything. Seems like standard marketing drivel to me.

well, that's your opinion. i think he said lots, and it was very informative, that's mine. marketing drivel doesn't have the words like "i" in it. ed s. is a good guy, actually uses these devices and answered the questions really well. keep in mind, microsoft usually doesn't do stuff like on community sites, let's encourage them and not do the trite things our communities tend to do like sit on the sidelines / peanut gallery and only offer up negative comments. or become like brighthand and have the big scoop like "THE XSCALE LIE" i'm not saying brighthand is going to do that, but it would be right up there with the "compaq update lie" and the "ugly side of enthusiasm"...i like this trend, microsoft is doing good here. dig it.

[Jonathan1]

It seems as if Intel is a bit surprised at how poorly it is performing with ARM4 code. MS had books and papers in 2000. big whoop. That isn't the same thing as a working product. of course MS knew about X-scale, but if they didn't have one until late 2001/early 2002, this is their fault how?

Do they REALLY need a working product to get a handle on its architecture? Program developers get their hands on Beta code of Microsoft OS's well in advance of the final RTM and still get a good idea of how to program for it. Don't you think that if Intel had even the beginnings of a "how to" for the X-Scale back in 2000/2001 that Microsoft should have started taking a serious look at it and allowed their development team to dive into the inner working of v5.

Microsoft should have been the one to initiate contact with Intel. I mean for god sake it is their product that’s on the line. Intel makes the hardware and then someone designs an OS around that hardware. That hardware isn't just catering to Microsoft. Admittedly Microsoft and its manufacturers will be using the vast majority of the X-Scale but not exclusively. Windows XP, until recently, didn’t support USB2. Is that Microsoft’s fault for not supporting that or Intel’s fault for not making it compatible with current drivers on XP?
The blame game can go either way. Is it the manufactures fault for using a CPU before it’s fully supported? Is it Intel’s fault for not catering to Microsoft? Is it Microsoft’s fault for not getting a patch that supports v5 of the architecture out in time to support its OEMs?

Personally I blame Microsoft. Why? They are the key focal point in all of this. Microsoft has to support the OEMs. Microsoft has to make the OS. Microsoft has to keep its OS fresh with current tech. The OEMs depend on Microsoft. Microsoft depends on Intel. Intel is just a hardware peddler. Is the hardware peddler supost to contact the client or the other way around?

The final question I have that won’t get answered is this: IF Microsoft went to Intel last spring would Intel have provided them with support for getting X-Scale optimization up and running? If yes then the blame can most assuredly be placed on the shoulders of Microsoft. If no then its Intel’s fault.

Not trying to get into an argument Ed. Just my .00000000000000002 cents on the subject. At the end of the day though blaming someone isn't going to change the fact that there is a problem here. :(

[mormondad]

Here's the bio of Ed Suwanjindar - a product manager executive for the Microsoft Mobility Group. He, apparently, is the PR lackey - no wonder his questions were so carefully, politically worded.


http://www.microsoftmobilitypress.com/site/extras/pics/thumbs/biopics/tb_v_mendillo.jpg
Ed Suwanjindar
Product Manager
Microsoft Mobility Group


As product manager in the Mobility Group, Ed Suwanjindar is responsible for marketing a range of new software for intelligent mobile computing devices. The Mobility Group is focused on developing software for targeted device categories, including the Pocket PC for PDAs; and the company’s upcoming Smartphone software.

Within the Mobility Group, Suwanjindar serves as the chief spokesperson for Microsoft’s mobile software products. As the primary public face for the group, Suwanjindar works with the product development, planning and business development teams to communicate Microsoft’s vision for mobility and product details to an audience that includes analysts, press and other industry influentials. In addition, he works closely with industry partners on joint marketing activities to create demand for Window-Powered mobile devices, particularly in the wireless space.

Before joining Microsoft, Suwanjindar managed marketing initiatives for consumer notebooks and devices at Dell Computer Corp. He graduated from University of Oregon.



I believe that we should be targetting our questions, comments, and complaints at the real decision-makers in the Mobility Group. You can see their names, positions, and faces at http://www.microsoftmobilitypress.com/site/home/main600.asp?SiteArea=600

You can leave Feedback for the Mobility Group at http://www.microsoftmobilitypress.com/site/home/main900.asp?SiteArea=900


I'm looking for email addresses to these executives now. We need to determine what exactly we want Microsoft to do about the problem (like release an XScale optimized version of PocketPC 2002 as a Flash Upgrade - among other things) and then convince them to do it.

[Foo Fighter]

Sadly enough, my iPAQ 3835 outperforms my Toshiba e740 in multitasking (including I.E. browsing...the e740 slows down significantly, the iPAQ does, but not remotely to the same extent). Same with basic program multi-tasking...The iPAQ outperforms in multimedia..

That is simply unforgivable, in my opinion. It sounds like going from bad to worse.

I was tempted to switch from Palm to PocketPC, but now I am losing all faith in PocketPC.

[jeffmckean]

Well, I think we can all be forgiven for a little disappointment in initial Xscale performance perceptions. Expectations were blown out of proportion, and Intel certainly has its share of the blame and in fact is not being accorded enough blame...after all, Intel certainly had the software available to test.

I think Ed was very brave for coming out to face the community. Ed is NOT a PR flack. He may do a lot of PR, but he uses these things and understands the ins and outs of all of them.

My feeling is that Microsoft has been very accessible to the community on these types of issues, and Ed, Derek, Beth and others often wind up being the point people. Give them credit for supporting the community. It may not always be what you want to hear (remember Derek saying that MSN Messenger and Terminal Services Client may not ever ship for RAM download? As I recall, that did wind up happening, and sooner than expected.) but it will be realistic.

It's an exciting time, lots of stuff in the pipeline. In a year, we won't remember what this was about.

[Paragon]

Wow! I just read the 'interview with Ed Suwanjindar and jumped to here to post so I haven't yet read everyones thoughts.... I just have to say that this guy definitely took bureaucratic bable 101. Sorry I don't mean to be rude, but he sure sounds like he's trying to make it sound like the Xscale processor, and the OS not working to well together is actually a good thing!!.

Kinda like when you get one of those letters from the city saying they are 'improving' the benefits you are receiving from them. After you go down the list you find you are getting less and they have raised your taxes, but somehow they make it sound like a good thing..... AAAAArrrrgggg.

Dave

[Newsboy]

Several thoughts...

Also glad I bought the Maestro for $209, instead of waiting.

This is why I got the Maestro, and gave up the EM-500, so I wouldn't have this problem down the road...*sigh*

Another classic case of companies working together using "documents", and "specifications". No human checks in place, so when your cow comes out looking like an elephant, everyone is standing around scratching their heads going huh? Then the lead engineer says, oh no problem, paint it white with brown spots. No one will notice.

BTW, I am a manufacturing engineer, I see stuff like this happen everyday.

[KH]

My experience in dealing with PMs at Microsoft is that most are fairly technically competent and also that they can never promise something that hasn't been formally blessed as a program. We'll see. I have been debating whether to cancel my IPAQ 3970 order but think I'll stay with it. I skipped the 37xx and 38xx, and am interested in an SDIO model that will still let me switch from naked to PC card according to my needs of the moment. I am very interested in knowing a few things though, if anyone has any knowledge.

First, and I think someone already asked this, are there any capabilities of the XSCALE that can be exploited by changes to the applications or to the development tools - or is an OS change mandated.

Second, there has been a lot of discussion about media-related applications, and indeed that interests me. But there are applications other than media or games that can benefit from performance boosts. Handwriting recognition programs like Calligrapher come to mind. Does anyone have relevant insight here?

Third, I have seen - but not seen corroborated - comments that without OS or app support the anticipated battery savings will not be realized. Does that apply to access of Bluetooth or SDIO facilities? Backlighting? Anything in particular?

Finally (for now) does anyone have knowledge about size of code differences between the ARM and XSCALE?

Kathryn

[igreen]

Interesting....we seem to be blaming Microsoft for the XScale performance issues? Why don't we apply the same logic to our PCs????? I mean shouldn't we expect a total rewrite of the Windows every time Intel does a speed bump or goes from Pentium 3 to Pentium 4? That would be ludicrous....I prefer not having to go buy new copies of the applications I use so they'll work on the "new" processor. No folks....this is Intel's problem. We've been able to see dramatic improvements from 2X MHz speed bumps on our desktop PCs without a total rewrite of the OS.

[ChrisD]

Remind me again, aren't you going to need higher MHz to run richer apps?
I think that would depend on the app. Games? Sure. Excel? Well, I have Excel XP on both a Pentium II-366 and a P-4-1.9GHz. I am a heavy Excel user and honestly cannot tell much difference unless I do something like run a big pivot table or load something with several thousand formulas.

I think ChrisD hit the nail on the head in many respects. The bus speed is the bottleneck. A 1GHz processor isn't going to do much until the bus is fatter and faster. :(

Thanks Ed! I did my homework on this one using the specs - even before XScale cpus were available to me. Everyone's experience just bears out what I was saying before. BTW, I think the benchmarks from Intel are misleading too. They are using a 64 bit memory bus so of course the mips rating is higer and the systems work faster.

[ChrisD]

Remind me again, aren't you going to need higher MHz to run richer apps?
I think that would depend on the app. Games? Sure. Excel? Well, I have Excel XP on both a Pentium II-366 and a P-4-1.9GHz. I am a heavy Excel user and honestly cannot tell much difference unless I do something like run a big pivot table or load something with several thousand formulas.

I think ChrisD hit the nail on the head in many respects. The bus speed is the bottleneck. A 1GHz processor isn't going to do much until the bus is fatter and faster. :(


Balogna! While I agree faster bus speed would help think about this Pentium 3's until recently were out perfroming Pentium 4's at nearly double the clock speed and with a much faster bus.... why? Software , the software does not take advantage of the 20 stage pipeline the P4's have. I don't blame Intel fully nor do I blame Microsoft fully, I blame them both for lack of communication that is the problem everyone wants to pass the buck. The bottom line is even with a faster bus I doubt we'd see much of any gain in performance.

There are major differences with the P4 cpu. Intel chose to focus on optimizing it a different way and that's why it's slower than the P3 like you mentioned.

[mormondad]

Interesting....we seem to be blaming Microsoft for the XScale performance issues? Why don't we apply the same logic to our PCs????? I mean shouldn't we expect a total rewrite of the Windows every time Intel does a speed bump or goes from Pentium 3 to Pentium 4? That would be ludicrous....I prefer not having to go buy new copies of the applications I use so they'll work on the "new" processor. No folks....this is Intel's problem. We've been able to see dramatic improvements from 2X MHz speed bumps on our desktop PCs without a total rewrite of the OS.

It is Microsoft that writes the Specs and requirements for these things. It's their baby. If the XScale was going to mess up the PocketPC, they should never have allowed it to happen under the PocketPC licensing. But it looks like no one even checked. Knowing that they had approved use of the XScale in upcoming PocketPCs, they should have been ready with properly written and optimized apps or patches to the OS. They should not have even let an OEM send a unit out the door without the underlying OS and apps being optimized to at least work properly.

Personally, I see lawsuits from OEMs and possibly a class action lawsuit over this. They were already sued and settled because of what was considered to be false advertising in some of their ads that didn't specifically state you had to buy the wireless hardware seperately. No reasonable person expected free wireless hardware, but they were forced to settle anyway. I see the XScale thing as more egregious and any reasonable person or OEM would expect to see a performance increase - not a significant decrease in some functions. The PocketPC platform may never recover from this blunder - it will generate bad press, bad publicity, bad feelings, bad sales, bad OEM relations, bad developer relations, and boatloads of just plain general badness. Andjust when the PocketPC was really starting to gain marketshare...

[Charles Pickrell]

It is my guess that Intel really expected Arm V4 apps to get some sort of a speed boost on the 400 MHz processor. Now that that is not the case, they are hoping everyone will optomize for their V5 processor. I read in Pocket PC magazine that Samsung will release an ARM 10 CPU that runs at 400-600 MHz and is expected to be very power efficient. Maybe that can be a holdover for the Pocket PC world until the OS moves to an ARM V5 version. Regarding my statement about not buying an X-Scale Pocket PC, let me revise that. If an X-Scale model comes out that better meets my needs I might consider the speed hit for extra features. I'm really wanting that Toshiba e550G from Japan with it's 4" screen.

All this 4, 5, and 10 stuff is getting me really confused! :)

Charles

[The PocketTV Team]

This blows.
Clearly PIM apps etc. won't benefit from the XScale, but can multimedia apps that would stand to gain from increased performance be written to take advantage of XScale, or will the OS prohibit that?

In other words, we know that PPC2002 doesn't and won't take advantage of XScale, but does that mean 3rd party apps can't either? Or can they be written that way at the expense of backwards compatibility?

Anyone?

Currently MS does not provide developers with any access to a C++ compiler capable of compiling Pocket PC application optimized for Xscale (i.e. instructions optimized for Xscale pipeline and availability of compiler intrinsics to use Xscale special instructions).

MS documentation indicates that those features are supported in eVC++ 4.0 (that's part of the .NET stuff), which sounds great.

Unfortunately the Pocket PC 2002 SDK requires eVC++ 3.0 and there seem to be no way to use eVC++ 4.0 for compiling Pocket PC applications, as far as we know.

Could anyone from MS confirm that ?

[Paragon]

The PocketPC platform may never recover from this blunder - it will generate bad press, bad publicity, bad feelings, bad sales, bad OEM relations, bad developer relations, and boatloads of just plain general badness. Andjust when the PocketPC was really starting to gain marketshare...

I hope this is not the case, but it sure is a possiblity. I posted a few weeks back that I thought it would be a better idea if everyone held of releasing xscale devices until everything was working together. To come out with devices that were not up to par would be a very difficult thing to come back from.....It's now beginning to look that way.

I think to counteract this manufacturers are going to have to focus on other hardware improvements, such as 802.11b, Bluetooth, better screens such as the new Ipaq has, GPRS, battery improvements, video chips, ect., and consider processors as a sideways move for a while??

Dave

[mormondad]

Currently MS does not provide developers with any access to a C++ compiler capable of compiling Pocket PC application optimized for Xscale (i.e. instructions optimized for Xscale pipeline and availability of compiler intrinsics to use Xscale special instructions).

MS documentation indicates that those features are supported in eVC++ 4.0 (that part of the .NET stuff), which sould great.

Unfortunately the Pocket PC 2002 SDK requires eVC++ 3.0 and there seem to be no way to use eVC++ 4.0 for compiling Pocket PC applications, as far as we know.

Could anyone from MS confirm that ?


So, now we hear from a PocketPC developer that Microsoft not only didn't pre-coordinate any of this with them, or send them the proper SDK so that at least THEY could release optimized apps for the XScale. But we find out that Microsoft hasn't even released a compiler that would allow a developer to optimze their apps.

The hits just keep on coming.

[The PocketTV Team]

SUWANJINDAR: “we don’t think that MHz is what ultimately matters to customers. What matters the most in this market is whether customers can do what they want to do with devices quickly and easily. With the richest set of software applications built into any PDA on the market, and the strong momentum that Pocket PC has with developers writing for our platform, we think that customers will be able to do the things they want to do with the performance they expect on devices using PXA250 processors.”

The problem is that the most popular class of applications on those devices is Video/Game/Multimedia, and it appears that Xscale devices are noticeable slower than StrongARM devices for those applications. Bummer!

We agree that MHz does not really matter, what count is well the software perform, how smooth, fluid and sharp the video is, etc. That's exactly the point. Here we get more MHz but less performances!!!

[ChrisD]

If you really want to see the XScale fly:

The following should be done by Microsoft:

1. Support Thumb (16 bit instruction set). This will reduce the code size about about 30% and this increases the speed since the extra 16 bits of code are not needed to be read each time. This requires a whole rewrite of the OS and all the applications.
2. Support ARM5 code. There are some optimizations included in the new instructions. This requires a whole rewrite of the OS and all applications.

The following should be done by the Manufacturer:

1. Support a 32 or 64 bit memory bus. The 64 bit bus is going to make the chip perform much faster since the cpu does not have to waste time reading 32 bit code and 32 bit data before performing the instruction.
2. Use a video chip like the Imageon 100 from ATI or the MediaQ chip. If you use a video chip, then provide end users with applications optimized for it - even if they are 3rd party. Examples of these applications include an MPEG 4 video player or a picture viewer. Without these apps the user will not see the whole benefit of using a video chip.

The following should be done by Intel:

1. Increase the depth of the 7 level pipelining to support full speed execution using a 16 bit memory bus for both read and write. This would require at least 9 levels of pipelining to do this. The first 4 cycles of the pipeline would be to just read the instruction and data!
2. Increase the internal cache. The cache runs at the cpu's clock speed and it is 64 bits so the load time is faster.
3. Add support for Cardbus. Right now the Xscale only supports PC Card slots. Without Cardbus, we will never get to use the full speed of a 100 megabit ethernet or other high speed connections like 802.11a.

Overall, I do not blame Microsoft for these issues. I do believe that the "HYPE" of 400 mhz is very misleading due to these issues. It's really something that Intel should be explaining to the consumer - just like they do with their x86 processors.

BTW, you can get all the details on the Xscale cpu at Intel's website - http://developer.intel.com/design/pca/prodbref/298620.htm i recommend reading the whitepaper to start with.

Also, I recommend reading the Intel® PXA250 and PXA210 Applications Processors Specification Updatehttp://developer.intel.com/design/pca/applicationsprocessors/specupdt/27853401.pdf to get the details on the bugs that are still in the processor.

[The PocketTV Team]

1. Support Thumb (16 bit instruction set). This will reduce the code size about about 30% and this increases the speed since the extra 16 bits of code are not needed to be read each time. This requires a whole rewrite of the OS and all the applications.

Or rather just a recompile of all the C/C++ code.

I'd be curious to know how much gain could be achieved by just recompiling the code with the /QRxscale option which should cause the scheduler to optimize for the XScale pipeline. Unfortunately the compiler used by the Pocket PC SDK does not support this option...

2. Support ARM5 code. There are some optimizations included in the new instructions. This requires a whole rewrite of the OS and all applications.

Probably not a full rewrite, but a rewrite of the critical routines that can take advantage of the additional instructions.

[Foo Fighter]

Yes Chris, but what are the chances of those things happening? Not likely anytime soon.

PocketPC has just been wrecked by the move to xScale, and no one at Microsoft or Intel seems to give a damn. Meanwhile PPC owners will be forced to live with crippled hardware, slow performance, and no relief in sight.

What is this going to do to PocketPC's growth? I think it just stopped here.

[The PocketTV Team]

Well, I'm sure that the most concerned people are not the users but the hardware manufacturers, and they'll find a way to solve the performance problem to some extent. They have a lot of money to loose if that does not happen.

[Skoobouy]

Wait! STOP! You all have it all wrong!

This has nothing to do with Microsoft or Intel. It's Toshiba's fault and has been from the beginning! Go over and read the Brighthand e740 forums.

1.) The Loox does not have the same multimedia performance problems as the e740, and it is an XScale PocketPC.

Quote from PocketTV Team: It looks like the Toshiba E-740 has some performance problems that are very significant compared to the LOOX (also running Xscale).

Video applications (not just PocketTV, also WMP and PocketDivX) are very, very slow on the Toshiba, while on the LOOX they run at almost the same speed as on Strong-ARM Pocket PCs.

We don't know precisely the nature of the problem on the Toshiba, and we don't know yet if Toshiba will be able to provide a software upgrade that will bring their device to par with the LOOX. But Toshiba will have to take some type of action if they want to keep their customers happy.

2.) The e740 has particular trouble moving memory, suggesting that most performance problems (including the multitasking one mentioned here) are memory related and not processor related.

Quote from PocketTV Team: The benchmark indicates that "memory move" is twice as slow on the e740 compared to StrongARM devices. This is an important point that was ignored in the PocketNow article.

This may indicate that the e740 performance problem is caused by slow memory access, rather than by problems with the video controler. And if that is the case, applications that read and write large amount of memory data, like video applications, would be significantly affectedby this problem, which is exactly what we observe.

3.) The e740 may also be having problems because it's the first Pocket PC to use a 3rd party graphics chip, namely the ATI Imageon chip; the performance problems might be based on a poor implementation by Toshiba.

Quote from ATI press release: "Toshiba PocketPC e740 users will benefit from the rich graphics solution provided by ATI's IMAGEON 100," said Oscar Koenders, vice president of worldwide product planning, Toshiba Computer Systems Group. "Higher frame rates and crisper displays, coupled with longer battery life, result in a better experience and productive usage for Toshiba PocketPC e740 users. ATI's leadership in quality graphics performance is a definite plus for Toshiba's new PDA."

4.) The e740 is still 30% faster than ARM Pocket PCs in all normally functioning areas! That's all you should expect. See: http://www.sprinklerhead.com/toshe740b.htm

5.) Once a program is optimized for XScale, it will no longer run on ARM machines.

Quote from PocketTV Team: There is a difference between "optimized for Xscale" and "compiled for Xscale".
> And would such "optimized" code run, or run acceptably, on ARM processors?

No, it would not run on STrongARM, unless the optimized portions of the code are duplicated, i.e. if the code detects the processor type and executes the optimized portions only when running on the Xscale.

It would be like what happened with MMX on x86 processors. MMX instructions do not exist on the Intel 386 processors.

> Will developers of performance-intensive software such as games and video playback now have to essentially program for two different processors, and offer two sets of code distributions?

Or just one set with the two versions of the critical routines. Usually more than 90% of the time is spent in less than 10% of the code. Only the very performance-critical routines would benefit of being optimized.

> Then two sets of patches for updates, etc? Sounds like that would defeat the purpose of having the Xscale be ARM compatible. Are we returning to the situation we had with MIPS, SH3, and ARM?

No, it is more like the introduction of MMX instructions in the Pentium.

Now, add two and two together: e740 is only working XScale unit to be effected + Problem may be memory related + Experimentary graphics chip + No other problems + Optimization is not a panacea = This is Toshiba's fault, so stop hanging Bill Gates in effigy and start complaining to the real cultprit! Toshiba is the one with the answers to our questions.

[bitbank]

This situation feels like deja-vu with the old MIPS compiler problem. 2.5 years ago I discovered that the MIPS compiler shipped with eVC++ was a piece of crap. It is buggy and created code which ran at about 1/2 speed on all MIPS CE devices. I raised this as a hot issue with WebTV and Casio since they were the biggest customers of the MIPS CPUs. BSQuared (the compiler author) was notified. It got brushed away never fixed. Now you have seen the results with the MIPS CPU being wiped off of the Windows CE CPU lineup. Again, technology loses to politics. As far as I'm concerned, I will write my applications to take advantage of the ARM5 instruction set by writing inline assembly language if no one provides a compiler with the support.

It is ridiculous for them to charge more money for a device which performs worse. Maybe some OEM should design a 300Mhz StrongARM device for those people who want a true high performance Pocket PC.

L.B.

[The PocketTV Team]

Yes, we all know that the e740 has a performance problem that is more serious than other Xscale devices.

But have you seen any other 400MHz Xscale Pocket PC running multimedia applications (e.g. PocketTV, WMP) faster than a 206 MHz StrongARM ?

We have not. The LOOX is almost as fast, but still a 5-10% slower than a (non-overclocked) iPaq on PocketTV.

[The PocketTV Team]

> 5.) Once a program is optimized for XScale, it will no longer run on ARM machines.

Yes, it's true (thanks for quoting us), but it is possible to include both the compatible code and the optimized code (for the critical routines) in the same binary, so that it would run on both platform by detecting dynamically the processot type.

The resulting code would just be a bit larger.

But since MS does not provide the Pocket PC development tools for compiling Xscale-optimised code, the question is hypothetical at this stage :)

[hulksmash]

I believe that despite all this very informative technical discussion that the positive aspects of this "problem" is that the pocket pc platform has gained some very real momentum from multiple manufacturers. Although hp/compaq control roughly 75% of the ppc 2k2 marketshare, competition is always a good thing, if the merger had not occurred, hp would have released the jornada 575 with wifi and the xscale processor instead of Toshiba, but since it DID happen, the hp ipaq will come out a bit later. It seems very possible that microsoft and intel are, for the moment, sidetracked from simply chasing the megahertz race to the much more exciting and potentially more impacting development of the pocketpc phone edition. The ppc 2k2 phone edition is currently under preliminary testing in Europe by hp and has a few scheduled releases in the US by the end of 2002. With the demise of so many of the wireless modem start-ups it makes sense that Microsoft is betting on pocketpc phone being much more lucrative in the long run, and is probably experiencing a fair amount of monopolistic salivating toward taking oversome of the very real profit margins and marketshare from Sprint, verizon, etc. All the previous posts, were, however really informative but I think they are moot for the short-term, but will hopefully be resolved soon. I was sitting on the edge of my seat, anticipating the xscale jornadas which were scheduled for release on May 15th, and although a few thousand were actually manufactured, they were immediately recalled with the finalization of the merger. I personally am glad I didn't buy one, with such lackluster claims to performance gain that they are destined to NOT have for the next few months, if not year and some odd months. :mrgreen:

[The PocketTV Team]

I was sitting on the edge of my seat, anticipating the xscale jornadas which were scheduled for release on May 15th, and although a few thousand were actually manufactured, they were immediately recalled with the finalization of the merger. I personally am glad I didn't buy one, with such lackluster claims to performance gain that they are destined to NOT have for the next few months, if not year and some odd months. :mrgreen:

I wish I had one! They're very valuable now (rare museum pieces). Would have been nice, next to our Jornada 420 (http://www.hp.com/cposupport/prodhome/hpjornada481202.html) prototype :)

[Equinox]

Doesnt this seem to indicate that Micro$oft has lost interest in the PPC ?

Remember what happened to the HPC ?

Microsoft had plenty of time and resources to develop PPC 2002 and ce.net for xscale - it hasnt which indicates to me that it sees the future of ce.net in devices other than the PPC

[mookie123]

From intel's white paper

Intel seems to emphasize 2 major additions in the Xscale:

-Tumb Code
-A MAC

Tumb code would improve memory utilization, and A MAC would speed up "Vital audio, video, and wireless applicaiont" with "dramatic" benefit.

Then there are V4, V5 specs differences which seems also to address performances in area that supposedly to speed up things like audio/video dramatically, instead of "same" or slower.

I really start to think It is Microsoft's blunder that cause the performance problem. (or marketing decission rather). All hardware documents seems to indicates that Intel and ARM puts considerable effort putting features to enhance multimedia speed.

I really fail to see why Microsoft insist that every future program would have to run on ALL StrongARM PPC2002 at the cost of sacrificing performance gain in Xscale.

[Timothy Rapson]

It's been 2+ years since the Ipaq 206MZ with PPC 2000?

In that time the desktop has gone from what, 366MZ Intel PIII to 2 GZ.

The Palm has gone from 8MZ to 16, to 33, to 66MZ and from no music, to MP3 and Atac, and from 160 by 160 graphics to 240 by 320, to 320 by 320 to 320 by 480. From no built-in camera to built-in camera.

Even the Mac has gone from Power Processor runnning 266 MZ to 800? (Sorry, I have not followed much of this.)

In all of these other systems, the increases were natural and progressive. It almost seems MS, the OEMs, and Intel went out of their way to mess this up. How could this happen?

How could it happen that in a closed system like the PPC where you don't even have to worry about an expansion bus, where you don't worry about internal memory expansion. Where the graphics, sound, and virtually all in/out standards are completly under the control of MS, you get no improvement at all?

I thought Palm was standing still in the middle of a cowpie. MS is tracking the stuff all over the carpet walking around in a circle.

[ctitanic]

2 things.

1-XScale does not mean 2x faster because of the bus factor.
2-I don't care about what MS is saying because 3er party companies will take care of creating applications which will use all the possibilities XScale is offering. Just to give one example: I have never used WMP, I use PocketDivx ;) The rest of 2002 in my opinion is performing well, but if somebody is offering me better performance in a product I wont think twice to don't use any more the MS product and start to use the new one. So.... Like I said, I don't care about what MS is saying.

[Equinox]

2 things.

1-XScale does not mean 2x faster because of the bus factor.
2-I don't care about what MS is saying because 3er party companies will take care of creating applications which will use all the possibilities XScale is offering. Just to give one example: I have never used WMP, I use PocketDivx ;) The rest of 2002 in my opinion is performing well, but if somebody is offering me better performance in a product I wont think twice to don't use any more the MS product and start to use the new one. So.... Like I said, I don't care about what MS is saying.

Youre missing the point - developers have to use microsofts API's and bits of code in order to construct their applications, if these are not optimised for the processor they are running on then they become a bottleneck slowing down the entire application.

[ctitanic]

2 things.

1-XScale does not mean 2x faster because of the bus factor.
2-I don't care about what MS is saying because 3er party companies will take care of creating applications which will use all the possibilities XScale is offering. Just to give one example: I have never used WMP, I use PocketDivx ;) The rest of 2002 in my opinion is performing well, but if somebody is offering me better performance in a product I wont think twice to don't use any more the MS product and start to use the new one. So.... Like I said, I don't care about what MS is saying.

Youre missing the point - developers have to use microsofts API's and bits of code in order to construct their applications, if these are not optimised for the processor they are running on then they become a bottleneck slowing down the entire application.

There are other ways around this problem. Have you forgot about the overclocking program for iPaq from Jimmysoftware? That's a clear example of what people can do. I contacted Mr. Castillo, one of the developer behind that application and his answer was "Yes, that can be done for XScale, and we know how to do it already"
So developers always will found the way around I don't have any doubt about.

Another point are "manufacturers" - MS mentioned iPaq users and defend iPaq users investments. But HP has released them new PPC and guess what, using XScale ;) Manufacturers will take care of MS. ;)

[marlof]

Another point are "manufacturers" - MS mentioned iPaq users and defend iPaq users investments. But HP has released them new PPC and guess what, using XScale ;) Manufacturers will take care of MS. ;)

Don't take things out of perspective. Ed S. talked about the installed base of 2 million current iPAQ users that don't have Xscale but "old-fashioned" StrongARM processors, and an upgrade path to Pocket PC 2002. Of course in future new processors will be used in different devices, and you can safely bet those will be supported, there's no mystery to that. The question is: "What is happening with Pocket PC 2002 and Xscale?" not "What will happen with Pocket PC 200X and future devices".

[Madoc Owain]

There's two different comparisons going on here, which is correct:

Compare the Xscale and ARM chips to Intel's 386 and 486 processors. Intel made the 486 backwards compatible with the 386, so people could continue to run their old software on it. Everyone could see an improvement in processing power due to the different speed and architecture, no tweaking needed to the OS to take advantage of this. Hell, if you wanted, you could even drop the clock speed back down to 8MHz for programs that depended on specific timing!

OR...

Compare the Xscale and ARM chips to Intel's Pentium and Pentium MMX processors. You'd see NO performance difference between the two, because applications had to be written to utilize the new MMX instruction set. If software were written specifically to use MMX then it would not run on the std. Pentium - I don't think this ever happened. Now, if the applications (this is pre-DirectX) began using MMX they'd see a speed increase, but would have to code for both non-MMX AND MMX processors for compatibility reasons.

I'm seeing XScale vs ARM as best fitting the second example. Now, what has happened since the early days of MMX is that the OS has taken over much if not all of the multimedia functions (DirectX in the PC, ??? in PPC) from the individual programs - for you youngsters out there, most high-performance games used to take over control of the hardware from the OS, causing some interesting problems at times but allowing them to run much faster because they didn't have to beg/compete for resources from/with the OS.

Along comes a marketing guy from MS, who tells us that MS will not be optimizing their code for XScale. Point by point, with translator microbes engaged, he says:

1. PPCs still outperform Palm so there's no need to optimize yet.
2. We're not changing to v5 optimization because we want to be able to continue supporting v4 users, all 2 million+ of them.
3. We're making small changes that won't break compatibility but don't think they're changes that will give any appreciable performance gains.
4. Mobile is a strategic business but we won't take advantage of better hardware that's available to us, thank you very much.
5. No one needs more than 640K of RAM.. I mean, performance at its current level is just fine, regardless of whether we can do better.

Ouch. My brain hurts from so much doublespeak translation.

To refute point #2, does M$ make any more money by updating PPC2002 for ARM v4 based devices. Nope, doesn't seem like it. Can present devices run Phone Edition? Nope again - well, POSSIBLY but the hardware is not available across a wide product range to do so. Did M$ care about compatibility when moving from PPC2000 to 2002? Good question - did they? Could everyone upgrade?

Refuting point #4, PPC2002 Phone Edition requires brand-new hardware, as its an implementation of integrated telephony and hand-held computing. THERE IS NO REASON NOT TO OPTIMIZE FOR XSCALE! Are there any PPC2002 Phone Edition products coming out that use ARM v4??? No? Didn't think so.

In addition to the above points, also consider M$'s long history of FUD, and how this goes completely against established behavorial norms for them. I am left to draw one of three conclusions:

1. M$ wants XScale to flop - wouldn't be the first time they sabotaged a product (*coughO/S2cough*)
2. It was the marketing guy's last day and he wanted to go out with a bang.
3. This is a stall tactic until .NET is fully realized.. more people likely to buy into their new product if they can say it runs +100% than their previous OS too.

If #3 is the case, M$ gets to sell PPC2002 Phone Edition and the OEMs get to make 2002 Phones w/ Xscale processors - M$ doesn't have to do any more work than they have to at this point, and can spend a product cycle debugging *maniacal laughter* erm working on their Xscale implementation, then release the new edition claiming greater performance. Everyone wins, because the major news carriers will NEVER pick up on this - they get too much advertising money from M$ to risk it. ZDNet = prime example.

It's so diabolical...

:twisted:

M.O.

[Jason Dunn]

I knew this thread would generate some serious discussion, but 65+ posts in less than 24 hours? Impressive. :-)

[Foo Fighter]

I knew this thread would generate some serious discussion, but 65+ posts in less than 24 hours? Impressive. :-)

Nothing gets those keystrokes flying like a little controversy. :P

[ChrisD]

Yes, we all know that the e740 has a performance problem that is more serious than other Xscale devices.

But have you seen any other 400MHz Xscale Pocket PC running multimedia applications (e.g. PocketTV, WMP) faster than a 206 MHz StrongARM ?

We have not. The LOOX is almost as fast, but still a 5-10% slower than a (non-overclocked) iPaq on PocketTV.

Well the 50 to 10% difference using XScale is probably due to the SLOWER memory bus of 100 mhz vs. the StrongARM at 103 mhz.

[Foo Fighter]

Well the 50 to 10% difference using XScale is probably due to the SLOWER memory bus of 100 mhz vs. the StrongARM at 103 mhz.

That may be true. But what hope does this give current xScale users? Doesn't this suggest that they will NEVER see a performance increase on their handheld? :?:

[CTSLICK]

I was sure that X-scale in conjunction with new hardware (like the e740)would really motivate me to upgrade. My EM-500 is a bit long in the tooth and I have to make concessions on what software I load, even with a good sized SD card. But X-scale looks like a bust for right now...I think I'll just sit out another round of PPC evolution.

Maybe I'll overclock my EM-500 for grins, or get another 16 Mb of memory added. Or maybe one of the "all-in-one" units on the horizon will be compelling enough. I hate waiting.

[Dan East]

Compare the Xscale and ARM chips to Intel's Pentium and Pentium MMX processors. You'd see NO performance difference between the two, because applications had to be written to utilize the new MMX instruction set.

That argument is only logical if we are comparing processors with like clock speed. One would expect to see roughly the same performance jump between a Pentium 166 to a Pentium MMX 300 as we should see between ARM 206 and XScale 400. We don't see it.

The point is that apps that perform a vast amount of raw, rudimentary math processing, like Pocket Quake, should experience a large jump in performance if the clock speed is indeed doubled. Even if the bus speed has not improved, PQ should see some increase in performance. At this point it runs slightly slower on the "faster" XScale hardware compared to 206 ARM.

Dan East

[/dev/niall]

Yeah, but... he didn't really say anything. I mean, replace a few nouns and he could be talking about anything. Seems like standard marketing drivel to me.

well, that's your opinion. i think he said lots, and it was very informative, that's mine. marketing drivel doesn't have the words like "i" in it. ed s. is a good guy, actually uses these devices and answered the questions really well. keep in mind, microsoft usually doesn't do stuff like on community sites, let's encourage them and not do the trite things our communities tend to do like sit on the sidelines / peanut gallery and only offer up negative comments.

Well, no offense, but I'm not a Pocket PC or Microsoft fanboy and I have no intention of pretending that he told me anything useful.

We are aware that PXA250 (XScale)-based devices are not demonstrating the huge performance gains that were anticipated. That said, Pocket PCs continue to offer the best performance and the richest functionality vs. other handhelds on the market today

Which tells me nothing new. Doesn't even hint at the reasons why.

...To that end we’ve worked to make our devices upgradeable. Moving to ARM V5 would break upgrade compatibility. We’re not prepared to strand an installed base of over 2 million iPAQ users.

Since I don't know this guy I would normally give him the benefit of the doubt and accept this answer as is. But, he works for Microsoft; so I don't see how they could claim to care about upgrade compatibility with a straight face.

THOUGHTS: Some industry analysts have said that Microsoft doesn't have any fix in place because Intel couldn't get the chips out in time.

SUWANJINDAR: “We have implemented and released specific software changes that our hardware partners are implementing without breaking compatibility for our OEMs and users. While we believe there may be incremental gains that could be had via small optimizations we are not convinced there are across the board improvements that would amount to any kind of dramatic system wide speed up. We have to develop software based on the processor architecture that offers the broadest compatibility for developers and when we shipped Pocket PC 2002 as it still is today, that was ARM V4.”

Hmm. Were the chips late? Did that affect Microsoft's development schedule? This answer translates to "Hey man, if we did nothing it would still work! We did nothing! It still works!". As to "we are not convinced there are across the board improvements that would amount to any kind of dramatic system wide speed up" all I can say is that anytime I've doubled my clock speed on any platform I expect to see across the board improvements. I can see them when I overclock my current Ipaq. Now before you go off explaining the differences between the architectures to me: I understand. Most consumers will not. While we're talking about expected performance... I'm not all that surprised about this. It's pretty obvious performance is not a priority in the mobile division. Upgrading my Ipaq to 2002 slowed it down considerably.

THOUGHTS: Some of those same analysts have said it will be 2004 until there's an OS that can use the XScale CPU properly. Is that an accurate estimate?

SUWANJINDAR: “It’s too early to talk about the next version of our software. That said, we’re committed to delivering best-in-class functionality and performance while providing a foundation that enables our developer community to continue to innovate and build successful businesses on our platform.

Microsoft considers mobile devices a strategic business. We are committed to working closely with Intel and other silicon vendors on delivering future versions of our Pocket PC and Smartphone devices. We have released specific software modifications to our OEMs that in total are all of the optimizations we believe are possible to maximize PXA250 performance (without causing incompatibilities for our OEMs and developers).”

Translation: We have no idea when we can do it. So why not say so? And don't pretend the reason you don't know is because of your commitment to performance, or developer communities.

In the mobile device space, we don’t think that MHz is what ultimately matters to customers. What matters the most in this market is whether customers can do what they want to do with devices quickly and easily. With the richest set of software applications built into any PDA on the market...

This sounds exactly like something a Palm executive would say when asked why their devices blow chunks at multimedia applications.

Look, I respect your opinion; I just happen to disagree. A lot. ;) I really couldn't care less if I never hear from or of this person again. I hold no animosity towards him.

[/dev/niall]

2 things.

1-XScale does not mean 2x faster because of the bus factor.


I'm seeing this a lot in the threads, and I really wish someone would back it up with some numbers. I know we shouldn't see a 2x performance leap.

A bus speed differences of 3mhz would not make that much of a difference on a desktop (though it would be quantifiable). Is it really that different on a Pocket PC?

[Scott R]

Wow, six pages. OK, some thoughts of mine:

1) I always knew it but I'll say it here. Geeks are a fickle bunch. Kind of off-topic but...MS made a bold and very successful move by embracing the geek community through a variety of ways. The problem is that it's kind of like playing with fire. Things were all rosy when MS had the highest-performing PDAs out there, but once the situation started to turn the loyal geek community is quick to jump ship. I say this not as a dig on the community here, but simply that despite this being a "PPC" site, the members are more technology fans than PPC fans.

2) Jason, I'm quite shocked by the tone of your questions to Ed. Wow. I think I would have dumbed down the questions a bit (any chance you can ask him some more simpler questions?). Here's what I would have asked:
a) Reports coming in are stating that XScale devices are actually performing slower than StrongARM devices. Other reports indicate that the PPC OS is not optimized for XScale, nor will it be for some time. Nevertheless, can a consumer buying into an XScale device expect that within the short term the speed problems will all be corrected so that, at the least, no process should run slower than when run on a StrongARM device?
b) Is it also reasonable to expect that within the short term, there will be an appreciable speed increase in some/most processes? Can you specify the types of processes which should show appreciable speed increases and those that won't?
c) How about power management? Is it reasonable to expect that within the short term, a consumer can expect any appreciable battery life improvements with an XScale-equipped PPC device?

3) I was wondering how long it would take before the PPC hardware vendors would reel MS in to reality. Creating an OS upgrade every 1.5 years which can't be utilized by a large percentage of its current users isn't going to cut it, at least not in the handheld market. An old desktop can be upgraded piecemeal to run the latest OS. A closed-hardware PDA cannot. MS started to move in this direction by standardizing on the StrongARM and mandating flash OS. Non-StrongARM PPC owners have been written off (sad, but necessary at this state). But HP/Compaq (from feedback from their corporate customers) has no doubt told MS quite clearly that they won't go along with another major OS revision which will obsolete the current StrongARM devices. With eVB shut down (and still buggy and feature-poor) and the Compact Framework still in beta, enterprise customers are probably already quite upset and impatient.

That all said, I don't think the future is completely dim. I have to imagine that there must be some hope of appreciable speed gains as a result of the XScale processor and that MS can introduce some OS tweaks that will improve performance further while not breaking backward-compatibility. As it stands, the PocketNow review revealed many speed figures showing appreciable speed gains (I think it's unrealistic for anyone to expect twice the performance just because it's twice the MHz). They just need to correct the graphics numbers and memory move numbers, as I recall. More important to me (and many others) is appreciable battery life gains. The incentive is certainly there to make it happen, otherwise this will really blow up in their faces.

Scott

[crispeto]

I think Compaq/Hp, NEC, ASUS, Casio, Toshiba, Fujitsu and many others should be pitching a fit. There is no way I am going to buy an expensive PPC that can't outperform my Maestro. I won't be upgrading for a long time. Had the performance been greatly improved, I probably would have upgraded by Christmas. I suppose I'm disappointed but I'm also happy with what I have now and it looks like I'll be staying this way.

[ChrisD]

Well the 50 to 10% difference using XScale is probably due to the SLOWER memory bus of 100 mhz vs. the StrongARM at 103 mhz.

That may be true. But what hope does this give current xScale users? Doesn't this suggest that they will NEVER see a performance increase on their handheld? :?:

That's what I said before - no improvements if the application is memory bound.

Also, Palm's new OS is not immune Hardware is hardware here so if a Palm OEM uses it they will see the same issues.

[ChrisD]

2 things.

1-XScale does not mean 2x faster because of the bus factor.


I'm seeing this a lot in the threads, and I really wish someone would back it up with some numbers. I know we shouldn't see a 2x performance leap.

A bus speed differences of 3mhz would not make that much of a difference on a desktop (though it would be quantifiable). Is it really that different on a Pocket PC?

Yes the bus is that big of a deal on these devices. The concept is simple. For each cycle a RISC cpu can execute 1 instruction. Each instruction is composed of an op code (command) and data of 32 bits each totalling 64 bits. So with a 100 mhz bus, it is only possible to run at 400 mhz when reading from the cache. Once you switch to accessing ram, it will be 100 mhz - the same as the bus speed. Now since the systems implement a 16 bit memory bus, it takes TWICE as many reads as a 32 bit bus and FOUR TIMES the number of reads of a 64 bit bus. So with the pipelining optimized for 2 memory reads, the system has to waste 2 more 100 mhz clock cycles (for a total of four 100 mhz clock cycles) to read the instruction and data before it executes anything! The same is true for writing to the ram on the way out.

So you can see that the issue is the bus. It's all about math based on 2 to the Nth power.

[/dev/niall]

So you can see that the issue is the bus. It's all about math based on 2 to the Nth power.

Right. And increasing the bus speed by 3mhz translates to what kind of performance difference? I understand how it all works, I'm just wondering what the numbers are on these devices.

[PJE]

Looking at the Intel site the XScale PXA250 can use 16 or 32 bit wide memory...

What's the probability that Toshiba thought it was simply required to replace a StrongArm running in 16-bit mode with the new XScale processor for instant speed up.... WRONG!

Anyone know any XScale chips running in 32-bit mode?

Obviously this has implications on cost, PCB board space and PCB complexity.

Regards,

PJE

[CTSLICK]

Front page rant coming up tomorrow. Intel had better run. :twisted:

Did ya' call of the dogs or are you still sculpting a finely crafted broadside?

[Jason Dunn]

Did ya' call of the dogs or are you still sculpting a finely crafted broadside?

Actually, I emailed someone at Intel and I'm hoping to get a response back from them about an interview - I thought it was better to get all the information before pointing a finger at anyone. I know, I know, quite uncharacteristic for Thoughts, but I figured it was appropriate. :wink:

[ChrisD]

So you can see that the issue is the bus. It's all about math based on 2 to the Nth power.

Right. And increasing the bus speed by 3mhz translates to what kind of performance difference? I understand how it all works, I'm just wondering what the numbers are on these devices.

All I can say is that there is only one of the original Pocket PCs with a 32 bit memroy bus - the Casio EG-800. The iPAQ and all the rest of the devices are all 16 bit. This makes routing the traces much easier on the motherboards.

[TomB]

Jason all I can say is thanks! For whatever reason you are the only guy taking on the issue of lower performance. I have no idea what happened to the other boards, but they are either ignoring Toshiba completely or focusing on other aspects of the e740. You are the man!

Chris, I am always impressed with your writing and understanding of the hardware involved. Thanks for your info!

[/dev/niall]

All I can say is that there is only one of the original Pocket PCs with a 32 bit memroy bus - the Casio EG-800. The iPAQ and all the rest of the devices are all 16 bit. This makes routing the traces much easier on the motherboards.
I think you've lost me. Are we talking about width or speed? Aren't the new XScale PPC's also 16bit? If that is the case, then wouldn't that preclude width being the cause of the performance hit? And if so, does a 3mhz bus speed difference really explain performance drops of 10-50%?

I hope I'm not sounding pissy, because that's not my intention. ;) You're just jumping around too quickly for my tired brain to follow.

[Dan East]

Well, I think some other websites are so concerned with courting MS and the OEMs, and reaping from the resulting relationships, that they cannot be on the forefront of raising such issues. They will join in once the cat is out of the bag, and it is old news.

Here's an excerpt of a Pocket Matrix thread discussing some of the XScale performance issues. Digby has benchmarked and compared XScale performance to that of a StrongARM iPaq. In most cases the XScale is faster, except for memory reads.
http://forums.pocketmatrix.com/viewtopic.php?t=6170&start=60

Dan East

[LarDude]

In the interview, Ed Suwajindar said something about: "We're not prepared to strand an installed base of over 2 million iPAQ users."

Is this BS or what? What about AMD and their plans for a MIPS based a 400-500 MHz processor for CE? Didn't M$ also indicate that they would be working with AMD to help expedite this? (Unless they were talking about CE in general and not planning on targeting the PocketPC platform).

That being said, does anyone know anything about the status of AMD's effort in this arena?

Secondly, anybody know about linux efforts as related to Xscale?

[ChrisD]

All I can say is that there is only one of the original Pocket PCs with a 32 bit memroy bus - the Casio EG-800. The iPAQ and all the rest of the devices are all 16 bit. This makes routing the traces much easier on the motherboards.
I think you've lost me. Are we talking about width or speed? Aren't the new XScale PPC's also 16bit? If that is the case, then wouldn't that preclude width being the cause of the performance hit? And if so, does a 3mhz bus speed difference really explain performance drops of 10-50%?

I hope I'm not sounding pissy, because that's not my intention. ;) You're just jumping around too quickly for my tired brain to follow.

Overall speed is a combination of the bus width (# of bits) and the clock speed (mhz). So it comes down to how much you can move in a cycle. So a quick formula would be bus width time bus speed equals the overall bus transfer limit.

The XScale design allows OEMS to use 16 bit external memory. That's what the designs have been using in the StrongARM as well. They use 16 bits because the number of bus lines and the sensitivity to noise is reduced over 32 bits.

The 3mhz does not sound like much but it would result in a performance penalty. It's a small one - similar to what the LOOX is reporting.

As for some of the other issues, please read the errata document I linked to. There outstanding bugs that need to be fixed to solve some signifigant problems for PDA users.

[ChrisD]

Well, I think some other websites are so concerned with courting MS and the OEMs, and reaping from the resulting relationships, that they cannot be on the forefront of raising such issues. They will join in once the cat is out of the bag, and it is old news.

Here's an excerpt of a Pocket Matrix thread discussing some of the XScale performance issues. Digby has benchmarked and compared XScale performance to that of a StrongARM iPaq. In most cases the XScale is faster, except for memory reads.
http://forums.pocketmatrix.com/viewtopic.php?t=6170&start=60

Dan East

One of the things I would ask is whether or not the test data is larger than the cache. If it is not then the XScale will look faster.

[/dev/niall]

The 3mhz does not sound like much but it would result in a performance penalty. It's a small one - similar to what the LOOX is reporting.

As for some of the other issues, please read the errata document I linked to. There outstanding bugs that need to be fixed to solve some signifigant problems for PDA users.

So basically, you are speculating that the 3mhz bus difference is what's causing the 10-50% performance drops?

Sorry for harping. I just don't understand how a 3mhz bus difference (when the bus width is the same and clock speed is greater) results in such a drastic loss of performance. Nothing you are saying or linking to is explaining that.

Yes, bus speed matters. Not that much (10-50%), and not when clock speed is doubled.

[Jason Dunn]

In the interview, Ed Suwajindar said something about: "We're not prepared to strand an installed base of over 2 million iPAQ users." Is this BS or what?

Why is that BS? If Microsoft had somehow managed to make Pocket PC 2002 ARM5 based (or, rather, ARM5 only), it wouldn't have worked work any ARM4 apps, right? Hell, the OS wouldn't have INSTALLED on an ARM4 CPU like the StrongARM.

What about AMD and their plans for a MIPS based a 400-500 MHz processor for CE? Didn't M$ also indicate that they would be working with AMD to help expedite this? (Unless they were talking about CE in general and not planning on targeting the PocketPC platform).

:!: Bing. :!: Make sure you know what you're talking about before you pick up a stone to throw - Pocket PC OS is ARM-only, not MIPS. The AMD CPUs will be used in embedded devices, but not Pocket PCs.

I appreciate that everyone is frustrated about this, but don't let emotion make you say unfair things. 8)

[st63z]

/dev/niall,

If it helps, the PTV team said the Loox was roughly around 5-10% slower (in other words, still supposedly faster than the e740). I think you might've gotten the 50% figure in your head from one mistyped miscommunication or something (not that it was hard to figure out) :)

[st63z]

If Microsoft had somehow managed to make Pocket PC 2002 ARM5 based (or, rather, ARM5 only), it wouldn't have worked work any ARM4 apps, right? Hell, the OS wouldn't have INSTALLED on an ARM4 CPU like the StrongARM.


Would the code have been bloated to an unmanageable size if MS had implemented branching instructions around the specific critical, affected routines and added V5-specific code for those. Is this still possible as a future OS patch? Or, has MS already done all they can in this regard within practical limits (which may be what Ed S. alluded to?) and has admitted that it still doesn't translate into significant improvements for some reason?

Also, I seem to remember that MS had some plans to do a DirectX equivalent for the PPC (GAPI superset and successor), that would make it easier for multimedia developers to take advantage of the hardware -- is this in CE.NET? Just curious...

[/dev/niall]

/dev/niall,

If it helps, the PTV team said the Loox was roughly around 5-10% slower (in other words, still supposedly faster than the e740). I think you might've gotten the 50% figure in your head from one mistyped miscommunication or something (not that it was hard to figure out) :)

Bingo!!! Thanks St ;)

Chris, if you're reading this... it all makes sense now.

Thanks!

[LarDude]

In the interview, Ed Suwajindar said something about: "We're not prepared to strand an installed base of over 2 million iPAQ users." Is this BS or what?

Why is that BS? If Microsoft had somehow managed to make Pocket PC 2002 ARM5 based (or, rather, ARM5 only), it wouldn't have worked work any ARM4 apps, right? Hell, the OS wouldn't have INSTALLED on an ARM4 CPU like the StrongARM.

...etc...

I appreciate that everyone is frustrated about this, but don't let emotion make you say unfair things. 8)

Yes, yes...perhaps it was a little unfair...but only a little. They (M$) certainly didn't seem to mind "stranding" users when they adopted their ARM-only stance. (Yes, yes, I remember -- and even agree with -- many of their arguments for this). Nonetheless, their present stance just seems to smack of convenience rather than any real concern IMO. But my real concern (and I see that it was echoed in several posts here) is that they are starting to sound like Palm. They'd better tone down the spin-machine...otherwise we might all find ourselves talking about nothing but the "zen" of PocketPC in the future.

[pt]

so toshiba seems to me to be the one who is "at fault" here, assuming there's fault to be given. why? they make the product and they sell it.

no one forced them to release this product into the market.

if oems start releasing products based on hardware that microsoft hasn't promised -anything- for. i'm not sure why we can be miffed at intel or microsoft.

surely someone, somewhere at toshiba did some type of testing on the units but chose to still ship.

did ms -ever- say they're going to do an x-scale version of the pocket pc os for oems who want to use these chips -now-? nope. and if intel told these oems that it'll run all code as fast or faster than a 206mhz processor, then they didn't manage expectations.

[fmcpherson]

Has anyone seen any advertising or marketing information from Toshiba promoting performance gains due to the X-Scale processor?

[Speed Racer]

This is a quote from a Dreampages ad for the new iPaqs but I think it shows that resellers are going to be implying that there is a noticeable improvement in speed with the X-Scale processor.

"STANDARD FEATURES
- Based on the award winning iPAQ Pocket PC platform, the iPAQ 3970 features the fast new Intel X-Scale 400 Mhz processor and a brilliant transflective TFT display with 240x320 resolution and 65K colors, and integrated Bluetooth Wireless"

[Skoobouy]

Frank,

ATI's press release (http://www.ati.com/companyinfo/press/2002/4507.html) about their Imageon graphics chip for Pocket PC has whole paragraphs going on and on about how the e740 will have enhanced graphics, framerates, better battery life, etc. etc.

Truth is, Toshiba had every intent of making the e740 into a multimedia wonder. HP/Compaq did too, but they gave up on the Imageon chip when they weren't having any success with it. Toshiba should have given up on it, too.

[Ganyeon]

I think it's definately in Microsoft's interest to get the PocketPC running full tilt on XScale, otherwise Palm V6 may come along and eat it for breakfast.
I don't think Palm is V5 ready anymore than PPC is. Both are optimized for V4 ARM code. All of Palm's partners are producing ARM V4 chips anyway. They just announced they will run on the X-Scale, but the didn't do anything for that. V5 runs V4, it just runs slow, negating much of the speed gains. :(

Actually, Palm's OS5 DOES take full advantageof the XScale...
http://www.palminfocenter.com/view_Story.asp?ID=3676

[The PocketTV Team]

so toshiba seems to me to be the one who is "at fault" here, assuming there's fault to be given. why? they make the product and they sell it.
Actually Toshiba does not make the e740, they only sell it. The e740 is made by Compal, a Taiwanese company.

[AhuhX]

Actually, Palm's OS5 DOES take full advantageof the XScale...
http://www.palminfocenter.com/view_Story.asp?ID=3676

Well sorry but that article is a load of crap.

"It is in Intel's best interest to make the Palm OS run as well as possible "

Note to the simple-minded: The above is an OPINION. It's isn't a fact.

Backing it up with some gloating quote from Palm's VP of marketting doesn't make it any more factual either.

You should wait till you see some technical articles with some *actual facts* before making that claim. Relying on the PPC FUD factory that is P.I.C. isn't going to get you anywhere close to the reality of the situation...

[The PocketTV Team]

Actually, Palm's OS5 DOES take full advantageof the XScale...
http://www.palminfocenter.com/view_Story.asp?ID=3676

This article is full of BS.

For example, it says:

> Pocket PC 2002 isn't and therefore will be limited to 200 MHz. A version of that operating system that is optimized for Xscale may not be available until 2004.

That's just not true. We have tested Xscale-based Pocket PC that do run at both 200 and 400 MHz (on some devices, you can selected either or let the device picks the clock speed based on the load).

Of course, the problem is that the performances of the Xscale at 400 MHz are about the same as a StrongARM at 206 MHz (sometimes a bit less on applications that do lots of memory access), and the performances at 200 MHz are less (not necessarily half, that depends on what the application does). And that does not depend of the OS being used, so PALM OS on Xscale will be impacted as well.

Also, the performances of CPU-intensive applications (games, multimedia) are mostly unrelated to the OS performances itself, since almost 100% of the CPU cycles are spent executing the application's code.

We have now determined (with a high degree of certainety) that the performance issues observed with video applications is mostly due to slow memory access, and therefore optimizing the code for for Xscale and using the Xscale special instructions will provide very little relief.

Another interresting bit of information: we have been able to compile the entire PocketTV code using the MS ARM compiler from eVC4.0 with the /QRxscale option (which is supposed to cause the scheduler to optimize for the XScale pipeline). All the other optimization options were unchanged. The resulting code on Xscale is... about 5% SLOWER than the code compiled for StrongARM!!! I don't know if we should laugh or cry.

[Ed Hansberry]

Actually, Palm's OS5 DOES take full advantageof the XScale...
http://www.palminfocenter.com/view_Story.asp?ID=3676

The Silicon.com artcle that he references indicates that it cannot.
Having only recently ported to StrongARM, Microsoft's main rival Palm is not likely to release a genuinely Xscale-optimised operating system any time soon either.
And taking that marketing line from Gina Clark doesn't mean much.

You do understand that if it really is X-Scale optimized, it won't run on StrongARM and other ARM V4 processors. That is going to be a surprise to Motorola and Texas Instruments who thought they were going to sell their ARM V4 chips to Palm and Palm OS licensees. AFAIK, Intel is the only one with ARM V5 out for these types of devices.

[mookie123]

What PALM has on their plate:

-TI OMAP1510 Arm core is TI enhanced 925 which is ARM 9 (V4 ?) + TMS320C55x DSP

-Xscale use V5TE + (various Intel technologies, for eg. Intel media procesing Technology)

All in all the two chips have pretty large DSP functions differences. In fact, Arm.com listed Xscale as different "core" architecture. I seriously doubt Palm can "ultilize" all instructions on both chips, specially the DSP enhancement, and integrated them flawlessly in their OS. They practically have to write 2 different OS'es to fully support the different extensions from those 2 chips. IF Microsoft is lazy enough not to support the V5 + intel media enhancement, I am not sure about Palm's ambitions on supporting even more instructions from differing companies.

---------------
v5TE
In 1999, the ARMv5TE architecture introduced the ARM DSP instruction set extensions providing up to 70% speed up for audio DSP applications. Many systems require the flexibility of a microcontroller combined with the data-processing capability of a DSP, historically forcing designers to compromise performance or cost, or adopt complex multi processor strategies. The DSP instruction set extensions deliver enhanced 16-bit and 32-bit arithmetic capabilities in a single general purpose CPU, providing improved performance and flexibility.

-----------------------


TI OMAP 1510 description
http://www.ridgerun.com/products/dsp/tms320vc5471/
http://focus.ti.com/pdfs/vf/wireless/omap1510_bulltn.pdf

ARM CPU HP
http://www.arm.com/armtech/CPUs?OpenDocument

Intel Xscale
http://developer.intel.com/design/intelxscale/

[Jason Dunn]

Actually, Palm's OS5 DOES take full advantageof the XScale...
http://www.palminfocenter.com/view_Story.asp?ID=3676

Yeah, for all the good it will do for them - remember that Palm will NOT have any XScale development tools out for quite some time. In fact, I haven't heard anything about any dev tools being released until OS6 - Palm is assuming that everyone will simply run the old Palm apps in the emulation mode.

And nothing in that Palm Infocenter articles says that Palm OS5 will have ARM5 support - I'd venture to say that they coded for ARM4 support and the XScale will simply run the emulator in ARM4 mode, not ARM5.

I think if you go and read the entire thread, you'll find quite a few people putting the article in perspective - it's quite misleading.

[pdagal]

Here's another bummer. I just read through Intel's PXA250 Applications Processor whitepaper and their "Using SD and SDIO with the Intel PXA250 MMC Controller" document, and it seems the XScale has an integrated MMC controller (nice for hardware makers to need one less chip). However, like the SD implementation on the iPAQ 3800 series, it's an MMC controller than supports SD (and also SDIO, unlike the 3800) but runs a 1 bit data bus rather than the 4 SD is capable of. So SD will likely run at MMC speeds, as it does on the iPAQ 3800. *sigh*
Read for yourself at http://developer.intel.com/design/pca/prodbref/298620.htm
The docs are on the right sidebar.

[st63z]

I could've sworn I read somewhere (maybe pocketnow forum?) that the Toshibas have the full 4-bit SDIO bus. Is this incorrect, or if correct, unique only to the Toshiba (among all other upcoming PPCs)?

[pdagal]

If Toshiba does have a 4 bit bus then they're using an external controller rather than the one built into the XScale (which would be hard to imagine). So far, I can't find any info on Toshiba's bus. I'll try dropping them an email on Monday.

I could've sworn I read somewhere (maybe pocketnow forum?) that the Toshibas have the full 4-bit SDIO bus. Is this incorrect, or if correct, unique only to the Toshiba (among all other upcoming PPCs)?

[vik]

What a shame the Pocket PC doesn't use Java like the Sharp Zaurus PDA does? Then there would be no cross-processor issues and hardware manufacturers would be able to use a range of CPUs without the problems encountered with Xscale. All it takes is a new JVM and everything else runs as if by magic.&lt;P>
I'm off to check up on SuperWaba (http://www.superwaba.org), which is a PDA-sized open Java. It works on Palms (and under Java) but the Pocket PC beta comes out in a month.

[The PocketTV Team]

What a shame the Pocket PC doesn't use Java like the Sharp Zaurus PDA does? Then there would be no cross-processor issues and hardware manufacturers would be able to use a range of CPUs without the problems encountered with Xscale. All it takes is a new JVM and everything else runs as if by magic.&lt;P>
I'm off to check up on SuperWaba (http://www.superwaba.org), which is a PDA-sized open Java. It works on Palms (and under Java) but the Pocket PC beta comes out in a month.
That would not really solve the problems encountered by CPU-intensive applications like video and games. As you know, those applications must be compiled in native code, even on the Zaurus.

[cyruski]

OS5 runs on ARM but it isn’t any great performance boost because its not ARM optimized.
Wasn't the whole OS coded for ARM? I watched a video comparing quicksheet drawing a pie chart on OS4 and OS5, and the OS5 one was a lot faster.

Also, Palm's new OS is not immune Hardware is hardware here so if a Palm OEM uses it they will see the same issues.
Isn't OS5 supposed to be 'immune' to such issues? I mean the PalmOS ready program seems to allow OEM's to use any ARM-compatible chip they like with the OS.

You do understand that if it really is X-Scale optimized, it won't run on StrongARM and other ARM V4 processors. That is going to be a surprise to Motorola and Texas Instruments who thought they were going to sell their ARM V4 chips to Palm and Palm OS licensees. AFAIK, Intel is the only one with ARM V5 out for these types of devices.
Well, as the only ARM-native code in OS5 is the core of the OS and the emulator, so it doesn't seem like an issue to rewrite them. I think the that problem will start when OS6 supports ARM-native applications.

[dripstone]

Hello.

I've been working with the SA1110 for the last four or five years and just recently (last year or so) have been working with the Xscale. I'll admit that I haven't read all the replys but is there any mention of the "multimedia DSP" that Intel's documentation talks about?

I've run a number of comparisons between the SA1110 and the Xscale using a number of different operating systems (WINCE/LINUX/QNX) and in each situation the SA1110 always seems to come out ahead. After further investigation I found that none of the OS's were making use of the Xscale DSP. None of the OS's had implemented the proper context switching needed to support the multimedia dsp. When I ported a multimedia app to make use of the Xscale dps I did some performance gains, but definately nothing worth celebrating about.

[The PocketTV Team]

None of the OS's had implemented the proper context switching needed to support the multimedia dsp.
Ouch!

Are you saying that Pocket PC 2002 kernel does not correctly save the DSP registers when switching contexts on Xscale ?

If that's the case, that would definitely preclude applications from taking advantage of the DSP instruction...

Hey, if true, this information will save lots of wasted time to people who are considering optimizing their code to use those DSP instructions!

[tthiel]

Buy the new Sony Clie at 66 mhz and full screen. its better than any Pocket PC or Palm.

Sounds like we're screwed. Microsoft has no intention of "optimizing" the OS for xScale. :oops:

[Timothy Rapson]

Actually, Palm's OS5 DOES take full advantageof the XScale...
http://www.palminfocenter.com/view_Story.asp?ID=3676

Well sorry but that article is a load of crap.

"It is in Intel's best interest to make the Palm OS run as well as possible "

Note to the simple-minded: The above is an OPINION. It's isn't a fact.

Backing it up with some gloating quote from Palm's VP of marketting doesn't make it any more factual either.

You should wait till you see some technical articles with some *actual facts* before making that claim. Relying on the PPC FUD factory that is P.I.C. isn't going to get you anywhere close to the reality of the situation...



That artcle is not a load of crap.

The articles all over that have said for over a year that the X-Scales are going to deliver greater speeds and lower power consumption in PPCs are crap. That may or may not be true of the new Palm OS 5 devices. The normal development of new OSs and processors show that speed does increase and power use does drop. The reason this whole thing is news is that it failed. I would be much more surprised if Palm OS does not take advantage of the X-Scale power after watching this debacle.


If the new Palms out next Spring with X-Scales don't take advantage of their features, you can come back here and say the PIC article was crap.

[The PocketTV Team]

That artcle is not a load of crap.
Yes it is. Sorry if you are among the people who wrote it. This article makes false statements like "the Pocket PC OS can only work on 200 MHz Xscale", which is definitely not true. All the new Xscale Pocket PC devices are running the processor at 400 MHz.

[ChrisD]

I could've sworn I read somewhere (maybe pocketnow forum?) that the Toshibas have the full 4-bit SDIO bus. Is this incorrect, or if correct, unique only to the Toshiba (among all other upcoming PPCs)?

Nope! None of the portable implementations of SD use 4 bits from what I've heard.

[AhuhX]

That artcle is not a load of crap.

The articles all over that have said for over a year that the X-Scales are going to deliver greater speeds and lower power consumption in PPCs are crap.


Who's talking about those articles? We are talking about the P.I.C article claiming that Palm OS 5 is optimised for XScale?


That may or may not be true of the new Palm OS 5 devices.


Which is EXACTLY what I said. I said it takes more than OPINION that Intel will opimise their DAL for XScale (which I ponder would actually make any difference since they seem to just be talking from out their a**ess) and a quote from a Palm VP marketting drone to convince me that this is a fact.


The normal development of new OSs and processors show that speed does increase and power use does drop.


Have you actually read any of the comments previous to mine? This isn't the normal case at all in some circumstances. Sometimes the OS/programs have to be optimised for the new processors architecture to have any speed benefits. Case in points so far: P3 -> P4, MMX.


The reason this whole thing is news is that it failed. I would be much more surprised if Palm OS does not take advantage of the X-Scale power after watching this debacle.


What failed Timothy. XScale? The bus speed on the Toshiba? The OS? The memory used? As plenty of people have mentioned there is more than likely to be a variety of reasons the performance on these new devices is weaker than existing StrongArm units at certain tasks. The P.I.C. article is just flinging FUD at MS as they are inclined to do. We'll prolly end up seeing performance is weaker for a variety of reasons.


If the new Palms out next Spring with X-Scales don't take advantage of their features, you can come back here and say the PIC article was crap.

LOL, who are you to tell me what to do? Why should I wait till the Palm's come out to say the article is crap? Already mentioned is that the article states PPC's with XScale can only be clocked at 200Mhz which is of course utter nonsense.

Other than that I an ENTIRELY justified in saying the article is a load of crap when people use it to claim that Palm OS 5 is optimised for XScale, when the article simply doesn't have any facts to support that.

Sorry, I just havn't been successfully programmed by P.I.C yet to believe OS 5 is the be all and end all of OS's. I'll stick to facts for now.

[AhuhX]

Buy the new Sony Clie at 66 mhz and full screen. its better than any Pocket PC or Palm.

Sounds like we're screwed. Microsoft has no intention of "optimizing" the OS for xScale. :oops:

Do you actually have something practical to say on the topic, or just doing pointless trolling?

[cyruski]

when the article simply doesn't have any facts to support that

FACT: Palm has agreed on a deal with Asustek on its OS5 devices.

They are expected to run XScale. Do you think Palm is that stupid to relase a device with a processor not optimized for it, when there is the example of the topic of this very thread?

Also, today I browsed through PalmSource 2002 Developer Conference slides. One of them said if was indeed possible to have native ARM code in OS5, but the application has to have a 68k version of it inside. When the application is launched, the correct part can be ran according to the OS.


Already mentioned is that the article states PPC's with XScale can only be clocked at 200Mhz
The PIC article mentioned that the fastest optimised PPC's run at 206Mhz ARMV5 processors.

I just keep hoping that CE.Net will be like PalmOS5, supporting all the processors avalible while keeping compatibility.

[The PocketTV Team]

The PIC article mentioned that the fastest optimised PPC's run at 206Mhz ARMV5 processors.

No, that's not what the article (http://www.palminfocenter.com/view_Story.asp?ID=3676) says, re-read it.

The PIC article (http://www.palminfocenter.com/view_Story.asp?ID=3676) says that Pocket PC 2002 can only run Xscale at 200 MHz, not at 400MHz.

This is incorrect. Many of the new Xscale-based Pocket PC's that are coming out (or about to) are running Xscale at 400 MHz. We have a couple of production and pre-production units.

And we do have some production and pre-production units, so I can tell you that this is a fact!

[dripstone]

The PocketTV Team wrote:

"Are you saying that Pocket PC 2002 kernel does not correctly save the DSP registers when switching contexts on Xscale ?"

Pocket PC 2002 wasn't one of my test operating systems. The os's that I used was QNX Nutrino, Linux, and WinCE 3.0. Don't know much about the PC 2002 kernel.

Most times if your the only person/app using the dsp you might get away with using it, but there's no way to be sure that you are the only on using the dsp.

[igreen]

It is Microsoft that writes the Specs and requirements for these things. It's their baby. If the XScale was going to mess up the PocketPC, they should never have allowed it to happen under the PocketPC licensing. But it looks like no one even checked. Knowing that they had approved use of the XScale in upcoming PocketPCs, they should have been ready with properly written and optimized apps or patches to the OS. They should not have even let an OEM send a unit out the door without the underlying OS and apps being optimized to at least work properly.

In a word... you are incorrect. Microsoft does NOT write the specs and requirements of the processor, Intel does. Trust me on this...I work for a computer development company dealing with both Intel and Microsoft daily.

Personally, I see lawsuits from OEMs and possibly a class action lawsuit over this. They were already sued and settled because of what was considered to be false advertising in some of their ads that didn't specifically state you had to buy the wireless hardware seperately. No reasonable person expected free wireless hardware, but they were forced to settle anyway. I see the XScale thing as more egregious and any reasonable person or OEM would expect to see a performance increase - not a significant decrease in some functions. The PocketPC platform may never recover from this blunder - it will generate bad press, bad publicity, bad feelings, bad sales, bad OEM relations, bad developer relations, and boatloads of just plain general badness. Andjust when the PocketPC was really starting to gain marketshare...[/quote]


Again Microsoft didn't make the hardware, Intel did. There are hundreds of processors on the market all with different characteristics and performance. The Pocket PC prior to PPC 2002 supported numerous processors not just ARM. The HW vendors chose the processor. Lawsuits...probably not. On the hardware side there are way to many things that affect the performance of the device. Also....anyone who claims that Intel just "sprung" the Xscales on OEMs is fooling themselves. There is more to making a PocketPC work than just sticking the processor in. The HW companies would know about the performance issues before their products hit the market.

I do agree with you...it will have the effect of stunting the growth and acceptance of the PPC.

[AhuhX]

FACT: Palm has agreed on a deal with Asustek on its OS5 devices.


Yes I knew that so what? Toshiba, Fujisiu, Siemens are also all producing XScale devices. The sky is blue. The moon isn't made of cheese. More pointless facts that don't attribute anything valid to the conjecture that Palm OS 5 is XScale optimised or not.


They are expected to run XScale. Do you think Palm is that stupid to relase a device with a processor not optimized for it, when there is the example of the topic of this very thread?


I believe Palm couldnt even tie it's own shoelaces. They would wait for one of their OEM's or providors to do it for them. As the P.I.C. article themselves claim, they expect Intel to somehow use all the extended features of XScale while retaining compatability with the other manufacturers extended instuction sets. Utter rubbish IMHO, and not supported by any FACTS presented so far.


Also, today I browsed through PalmSource 2002 Developer Conference slides. One of them said if was indeed possible to have native ARM code in OS5, but the application has to have a 68k version of it inside. When the application is launched, the correct part can be ran according to the OS.


Again, nothing I don't already know. The fact that programmers can run native ARM by coding outside of the emulator environment has been brought up long ago. Unfortuntley it doesn't do much for the argument that PALM OS 5 is optimised for XScale. It has ABSOLUTLEY NOTHING to do with it.


The PIC article mentioned that the fastest optimised PPC's run at 206Mhz ARMV5 processors.


Well PocketTV has already corrected you on this, but they clearly state that Pocket PC's can only be clocked at 200mhz for XScale. Perhaps the most FUD-tastic statement I've ever had the misfortune to read on P.I.C.

The scary thing is some people actually believe it.

[AhuhX]

FYI, Just found this quote off SlashDot. Anyone of you jokers claiming OS 5 is optimised for XScale want to refute it? :twisted:

Pocket PC development uses Microsoft's ARM compiler. Palm OS development for OS 5.0 requires ARM Ltd's ADS 1.X tool chain, which likewise is not optimized for XScale.

I guess you get your OS 5 code optimised for XScale by clicking your heals together three times and repeating "It is optimised, It is optimised, It is optimised..."

Now back to topic... This guy posted some really interesting info...


It's important to differentiate between architecture optimizations
and CPU specific optimizations. The ARMv5 instruction set is a
relatively minor architectural tweak to the ARMv4 instruction set.
The names give you the impression that it's some grand change between
v4 and v5, if a technical guy did the naming it would be ARMv4 and
ARMv4.01. ARM is playing some games with architecture naming
to protect their business position with patents in a silly way.

ARMv5 adds a couple of new instructions over v4, an instruction to count
leading zeros in a register (which a compiler would likely never
use), and a better method of switching between the ARM instruction
set and the 16-bit Thumb instruction set. The later isn't
relevant for PocketPC since Thumb mode isn't supported. I think
v5 might having a new debugging hook as well.

The new XScale parts are ARMv5te, the T is for the 16-bit Thumb
instruction set, which no one seems to care about. The "E" adds
some DSP oriented instructions that are pretty interesting for
media codecs and such. They are the MMX equivalent for the ARM
world. They likely won't improve performance of the general
purpose aspects of the platform.

I think it's a red herring to chase Microsoft for not optimizing for
the ARMv5, the changes are really small and I don't see any
performance impact, certainly not if you have to maintain another
version for all of the strongARM based products.

Now, as far as CPU specific optimizations for the PXA250 (XScale)
implementation of the ARM architecture. IMHO Intel chased
MHz and left behind a lot of good sense about system performance.
The high order bit is bus performance as others have already
pointed out.

In addition to the bus performance, Intel made many tradeoffs
to optimize for clock speed: The 7-stage pipe has a 4-clock penalty
for a mis-predicted branch. This is compared to the circuit
design heroics in the strongARM that implements "all branches
are 2-cycles". The Xscale approach is much more complicated, it
probably doesn't perform any better, but you get a high clock speed.

Intel adds clock cycles to all load/store-multiple instructions
in Xscale. This is a pretty big deal in ARM since they are
used in the entry and exit of most C functions, in memcpy(),
and any time you are moving chunks bigger than a register.

The load-use penalty is bigger in Xscale. This is a pretty big
deal in ARM. The ARM instruction set is pretty compact. It is a
RISC processor, but the combination of shifting operations
combined with ALU operations makes it possible for a good compiler
to generate reasonably compact code. As a result, it's harder
for a compiler to put instructions between a load and instructions
that use the destination of the load. This is another trade-off
in Xscale that allows a higher clock speed but hurts performance
otherwise.

I go on too long, but the DEC designed strongARM used in the SA1100
is a tour-de-force of clean implementation and balanced system
performance. It's amazing that core was designed in 1993 (I think,
someone please correct me) and is still the leader for handheld
apps. The Intel guys went after clock speed at the expense of
everything else in Xscale and it will probably never optimize well
for a platform like PocketPC.

jeff


More at:
http://slashdot.org/articles/02/06/22/185245.shtml?tid=100

[argent]

Regardless of the toolchain in use, you can always optimise the assembly. If the XScale has a bigger misprediction penalty, then you need to boost tests a few instructions earlier in the code. If you can't do that, then you want to use mathematical rather than branch operations to get the same result. Most processors with a branch misprediction penalty have some way of indicating the most likely branch. This can be done automagically in a post-processing step... it should even be possible to write a recompiler that can do it.

Even in higher level code, you can unroll loops by hand to avoid branches.

Finally, on the Palm at least there's a variety of toolchains available.

Hopefully Intel will get some of those ex-DEC developers into the loop for the next iteration, after they finish bailing out Itanic. :wink:

[RobertCF]

In answer to the question about why manufacturers are busy pushing out the new XScale-based devices, the simple and obvious answer is that those are the chips that are being made. Regardless of whether or not ANY OS maker is optimizing their code for the chip, it's the successor to the 206Mhz chip, plain and simple. Since it maintains the backward compatibility, all the manufacturers know at the very least they DON'T have to change a darned thing in their software to produce a PocketPC device (or Palm, now, I guess). I mean, come on, what choice do they have---the old 206Mhz chip is essentially over and done with from an OEM stand point.

The real question and concern, which have been raised already, is when someone is going to throw resources into optimizing for the newer chip. And, as an owner of the original 3630, currently running PPC2k2, I've not been impressed enough with the new devices to shell out $700. I've never had a single problem with my 3630 and all OS patches and upgrades have gone in flawlessly. I don't have the dust or other problems some folks have run into. So, I might think about buying a new device IF mine finally dies or the prices on the new ones come down to about $400 (how many "s" in "Fat Chance"?).

[argent]

Regardless of whether or not ANY OS maker is optimizing their code for the chip, it's the successor to the 206Mhz chip, plain and simple.
How do you figure? There's a dozen ARM implementations from different companies... why is XScale the successor to the StrongARM any more than OMAP or DragonBall? Just because it's by Intel? Hey, my PCs have gone from Intel to Cyrix, back to Intel, to AMD, and who knows what the next one will run.

Xscale is being used because it was expected to be significantly faster than StrongARM. If it turns out that it's slower in practice, they won't keep using it. If the expected performance and power improvements don't eventuate, then they'll go back to the StrongARM. Or even the DragonBall, with its hardware MPEG coprocessor...

[The PocketTV Team]

Regardless of the toolchain in use, you can always optimise the assembly. If the XScale has a bigger misprediction penalty, then you need to boost tests a few instructions earlier in the code. If you can't do that, then you want to use mathematical rather than branch operations to get the same result. Most processors with a branch misprediction penalty have some way of indicating the most likely branch. This can be done automagically in a post-processing step... it should even be possible to write a recompiler that can do it.
I was gonna say the same thing.

It would be nice to have a compiler that can recompile and generate better Xscale code by using profiling data from running the application.

[Chazzta]

So you can see that the issue is the bus. It's all about math based on 2 to the Nth power.

Right. And increasing the bus speed by 3mhz translates to what kind of performance difference? I understand how it all works, I'm just wondering what the numbers are on these devices.

All I can say is that there is only one of the original Pocket PCs with a 32 bit memroy bus - the Casio EG-800. The iPAQ and all the rest of the devices are all 16 bit. This makes routing the traces much easier on the motherboards.

According to the iPAQ specs at http://handhelds.org/Compaq/iPAQH3600/iPAQ_H3600.html, the SDRAM bus is 32 bits, not 16.

1.4. SDRAM
The iPAQ H3600 contains 2 SDRAM chips configured as a 32 bit wide memory bus. The SA-1110 can support up to 256Mb SDRAM chips. The 64/128/256 Mb SDRAM pin assignment must be compatible with the iPAQ H3600.

[jjoe]

Wow, Intel puts out a chip that, when running the EXACT same software, runs it slower (per megahertz and absolute), and you guys are all blaming Microsoft.

think about it. If AMD came out with a new 4Ghz chip that ran Windows as fast as a PII-400 wouldn't you bash AMD.

I guess Intel propaganda really works.

By the way, if you think about ChrisD's comment about the memory bus being slow, XScale's ability to calculate Multimedia codecs may blaze, but if it can't pull data from memory fast enough because it's memory bus is dogmeat, then XScale will still sit there twiddling it's thumbs waiting for it's slow memory bus to feed it data.

don't drink the koolaid till you stop and ask what's it's spiked with.

[Chazzta]

Regardless of the toolchain in use, you can always optimise the assembly. If the XScale has a bigger misprediction penalty, then you need to boost tests a few instructions earlier in the code. If you can't do that, then you want to use mathematical rather than branch operations to get the same result. Most processors with a branch misprediction penalty have some way of indicating the most likely branch. This can be done automagically in a post-processing step... it should even be possible to write a recompiler that can do it.
I was gonna say the same thing.

It would be nice to have a compiler that can recompile and generate better Xscale code by using profiling data from running the application.

From the Intel® XScale™ Microarchitecture for the PXA250 and PXA210 Applications Processors User's Manual (http://www.intel.com/design/pca/applicationsprocessors/manuals/index.htm):

Branch Target Buffer 5
The Intel® XScale™ core uses dynamic branch prediction to reduce the penalties associated with
changing the flow of program execution. The Intel® XScale™ core features a branch target buffer
that provides the instruction cache with the target address of branch type instructions. The branch
target buffer is implemented as a 128-entry, direct mapped cache.

[cyruski]

The PIC article says that Pocket PC 2002 can only run Xscale at 200 MHz, not at 400MHz.

You can understand whatever you want from a text by interpreting single sentences 8) I clearly remember a thing called "grasping the meaning from the context" from my English lessons..

Read the second sentence, please:
Pocket PC 2002 isn't and therefore will be limited to 200 MHz. A version of that operating system that is optimizedfor Xscale may not be available until 2004.

[cyruski]

Yes I knew that so what?
Again, nothing I don't already know.

Just a single word, take no offense pls:
know-it-all :wink:

[AhuhX]

Read the second sentence, please:

Pocket PC 2002 isn't and therefore will be limited to 200 MHz. A version of that operating system that is optimizedfor Xscale may not be available until 2004.

Ahuh.. and....? The two sentences reveal two completley individual and distinct "facts".

1.) That the Pocket PC OS is limited to 200Mhz &lt;- Baloney
2.) An optimised version of the OS isn't due possibly till 2004 &lt;- Which may or may not be true.

Now, number (1) is crap and a complete mis-truth no matter how you look at it, whether the PPC OS is optimised in 2004 for XScale or not.

[AhuhX]

Yes I knew that so what?
Again, nothing I don't already know.

Just a single word, take no offense pls:
know-it-all :wink:

LOL, hardly as argent and PocketTV have demonstrated hehehe. I'm really not a technical person. The lowest I go is C, but I just find the whole discussion on the differences in architecture and design that Intel has taken between the two chips fascinating.

Sorry, but you really wern't adding anything to that article to convince me that OS 5 is optimised or not. Not that I really care, since it's an aside to a far more interesting discussion anyway. ;)

I mean, I've actually found the discussions on the Linux side of the fence (through SlashDot etc) far more interesting because the better posters there seem to know what they are talking about, are open-minded about the issue, and CLEARY state with technical facts why things are the way they are. It's almost like they are stepping back and looking at the issue as is because they arn't currently as close to it (and heated) as some PPC'ers are, or using it as some silly rant againt MS like the jokers at P.I.C.

At least I feel I'm actually learning something from them. With the P.I.C article it felt like... please leave your brain at the door while we wind the FUD machine up.

Anyway, I'd be very interested to know more technical details about how OS 5 tackles this issue, but somehow I don't think I'm going to find it at Palm Info Centre. :)

If you can point me to some links with the same level of factual info as some of the ones I've seen on the PPC and Linux sites I'd be more than happy to look at them.

Cheers

[pdagal]

My contact at Toshiba said that it is indeed a 4 bit bus! How cool and surprising. :o

If Toshiba does have a 4 bit bus then they're using an external controller rather than the one built into the XScale (which would be hard to imagine). So far, I can't find any info on Toshiba's bus. I'll try dropping them an email on Monday.

[digital-doc]

This MS response is a paragon of backpedaling corporate spin, aka BS.
Palm should use it for marketing in IMO.

[the white jb]

i develop software for pocket pc's and am trying to figure out why my app does not run properly on any pocket pc with an xscale processor. the app has been built for almost every 3.0/processor combination and runs just fine on a 2002 device with a strong arm processor. but on an xscale device, the app starts ok and then after a while the screen does not redraw properly and i get lots of crashes. anyone else experience something similar to this? or have any info on where i might be able to read up on some development issues related to xscale on the pocket pc?

any and all info would be greatly appreciated!

thanks!
chris

[The PocketTV Team]

Normally there should not be any problem like that as the Xscale is 100% compatible with the StrongARM instruction set.

So all StrongARM should run (with variable performances) but they should run fine. We have not heard of any problem like the one you describe.

Do you observe the exact problem with several different Xscale devices ?

A possibility would be that your program has a bug that was not triggered with ARM. Xscale being different, execution timings are different, and if your program is multithreaded, it is very possible (and even likely) that the thread will not synchronize the same way, because they will run at different speed an an Xscale than on a StrongARM.

So this may trigger hidden bugs there where present but not triggered earlier.

[the white jb]

yeah, i kind of figured that some bug would be the cause of the problem. i do know that we are having problems on the ipaq 3835 and the toshiba e740, but i will have to see if they are the same problem and try to reproduce them and debug them. i tracked down one bug on the e740 to the windows call 'Polyline'. i will have to see if i can reproduce this on the ipaq.

thanks for the info pockettv!

chris...

[The PocketTV Team]

> we are having problems on the ipaq 3835

This is a StrongARM device, not Xscale...

[the white jb]

yeah, sorry about that...i realized my mistake after i made the post...i thought that i still had the xscale, but someone hijacked it from me and gave me an old device instead...damn xscale junkies! :P

[The PocketTV Team]

If you have a problem with 'Polyline', it is very possible that this is caused by a bug in the video driver of the e740.

As you know, the e740 uses an ATI Imageon graphic controler, so most likely the graphic driver are taking advantage of some hardware acceleration in the Imageon. It is very possible that they made mistake in their driver. That would have nothing to do with Xscale, but rather with drivers on specific devices (e.g. e740).

If you can reproduce the problem and can write a precise bug report that can be reproduced, you should submit it to the Toshiba support (or send us the report, we can forward it to the right peopole at Toshiba).