Need For Speed: ARM Multi-Core Processors Coming Soon

[Jonathon Watkins]

<div class='os_post_top_link'><a href='http://techreport.com' target='_blank'>http://techreport.com</a><br /><br /></div>Spotted this at Tech Report: " It seems that everyone is getting into multi-core chips these days. ARM has unveiled a multi-processor chip that can incorporate as many as four processor cores. The "MPCore" is designed to run between 335 and 550MHz and is targeted at embedded applications, so don't expect one on your desktop. However, chips based on the MPCore are expected out in the first half of next year and could make their way into cars, consumer electronics devices, and even PDAs."<br /><br />SMP on a PDA? I know we've joked about it in the past, but, wow. 8O So who needs *that* much power on their PDA? :wink: Go on, I'm curious, what do you guys do with your PDA that could really benefit from multiple processors?<br /><br />For a suitably detailed insight into the Processors (including a diagram of the MPCore architecture) you can visit <a href="http://www.linuxdevices.com/news/NS3610443018.html">here</a>, or the <a href="http://www.arm.com/products/CPUs/MPCoreMultiprocessor.html">ARM website</a> itself.

[bjornkeizers]

the answer to that is easy.. I need that kind of power because I can have that kind of power. why would I settle for less?

[Pony99CA]

the answer to that is easy.. I need that kind of power because I can have that kind of power.
I think the correct term is "W? BIC!" :-)

why would I settle for less?
Ummm, because you don't want to pay for it? Why do any of us settle for less than we really want?

Steve

[R K]

Maybe because multiple processors would help to give me a better experience playing "Need for Speed" on a Pocket PC?

At first glance, I thought your post was about Need for Speed coming to ARM processors.

All kidding aside, I'm sure multiple CPUs would give a great benefit to multimedia applications if nothing else. Although you can say that it might be a waste to dedicated entire CPUs to seperate parts of a multimedia app when you can use DSPs and GPUs to do it more efficiently?

[baralong]

ummm how about games :snipersmile:

Voice recognition, Better/faster handwriting recognition, Multimedia

And perhaps route finding (http://www.jenneth.info/archives/003490.html, might be processor bound)

Perhaps distributed computation 0X SETI (http://setiathome.ssl.berkeley.edu)

Compression.

Probably it'll be a case of "Build it and they will come"

&lt;edit>Might it also be possible that if on of he new processors uses a similar amount of power as an old one we'd beable to underclock and get it to use less power?&lt;/edit>

[Jorlin]

Hmmm... playing DOOM3 on your PPC might be nice... wouldn't you agree.. :lol:

[R K]

Not if the speed's going to be measured in Seconds-Per-Frame instead of Frames-Per-Second.

[R K]

Hey buddy, what's the framerate you're getting on your Pocket PC?

I'M GETTING A WHOLE 87 SPF MAN!!!

:oops: OOPS... That's a bad thing. :oops:

[epdm2be]

Hi,

Contrary to what Mr. Watkins believes. These chips may make it in desktop systems. See http://www.iyonix.com for more details. RISC OS users might know what this is.

Also further on that site see the news page and read about Merlin (or that companie's plans for future RISC OS). You'll see clearly that they're devoted to the desktop market of that OS. Which is coincidentally tied to ARM processors.

In fact I think it would be exciting to use these cpu's in a full desktop environment. Low power and yet fast enough for most applications (as proven by PPCs).

Regards,

EPDM

[Jonathon Watkins]

Hi,

Contrary to what Mr. Watkins believes. These chips may make it in desktop systems.

:?: Pardon? I was quoting an atricle. It's not what *I* think.

(You can tell it's a quote by the quote marks). :wink:

I'm all for "W? BIC!", but I was asked exactly *how* all this power (with extra cost) would be most usefull for you guys.

[Felix Torres]

Multicores are a way to increase throughput without increasing clock-rate.
So, instead of doing a 1GHz ARM-9, you can substitute a dual-core 500 mhz system.

Of course, this presupposes the existence of (head for the hills!) a fully multi-threaded pre-emptive multitasking OS and apps. :twisted:

Yes, *that* old buggaboo...

BTW, notice I didn't say symmetric multitasking? thats cause for PDAs it might not be the best way to go.
On a PDA, dual cores would let you assign one cpu to background tasks and another to the foreground tasks (desktop Windows allows end-user choice of symmetric or asymmetric multiprocessing, btw, by letting you assign one cpu exclusively to one task) so music playback or communications functions can be managed by one core while the other handles user interactions.
Another advantage of multicores is that by not timeslicing a single high-speed cpu to run multiple threads, you don't need the higher clock-rates and the attendant high memory-bus multipliers, which generally mean wait states.
(Anybody remember the first 400Hz strongarms?)
So you can use cheaper RAM.
A dual 500MHz CPU with 100 MHz RAM only needs a multiplier of 5 instead of a multiplier of ten. Plus, you gain an extra level of control over power consumption by being able to disable one core completely when its not running at a high enough load.

The third advantage is in context-switching.
Every time a multitasking OS switches from one thread to another, it has to copy out the contents of the registers and working RAM to a safe location before it can load in the data for the new thread. As the CPU juggles multiple active threads, switching these various contexts back and forth starts to consume significant chunks of time. The same is true, but at even greater cost, for non-multitasking OSes doing application switching to simmulate actual multitasking since they have to copy out the whole application and its memory footprint, shut it down gracefully, and then restart the new one, over and over.
And, even worst of all are the heap-based partitioned-memory architectures where the whole heap has to be copied in and out. Thats why nobody ever got multi-tasking going on the old VIC-20s, the (I think) last heap-based architecture before PALMOS...

Anyway, a multicore system doesn't need to switch context as often as a single-threaded cpu running multitasking and that is a *major* gain in performance when running multiple multithreaded apps.

By the way, there is another level of hardware multitasking in between single core and dual-core cpus; the superscalar or hyperthreaded CPUs where the various execution units (integer, floating-point, MMC, etc) can be scheduled independently with different program threads. The P4 does this best so that something like 15% of the time a single CPU looks like a dual cpu to the OS.

So, for the near future, systems are likely going to be characterized by cpu sockets and the total number of execution threads the OS can see; for example, a dual-core hyperthreaded cpu would be a one-socket, four thread cpu, but a dual core non-hyperthreaded cpu would be a one-socket two thread system.

Just what we needed, right?
More technical complexity.
The thing to keep in mind is that multi-threaded hardware only provides benefits if the software is multithreaded or there is serious multi-tasking going on.
On the desktop the gains are way less than the theoretical maximum because most codes are singlethreaded and user-bound these days so most desktpop cpus rerally spend the bulk of their time twiddling their thumbs while the bio-ware reads the screen and makes up its mind which link to click on in the browser. ;-)

Oddly enough, there are two areas where portable multithreading would be very useful; gaming and media players. Expect the dual cores to be featured prominently in future PMC boxes.

Oh, nice to see the PDA hardware finally catching up to the arcitectural capabilities of WinCE. Not sure if the hal and kernel need tweaking for multiprocessors but the high-level APIs are already multithreaded so it won't be hard to get CE going on those future multicore systems.

I guess we could say that one person's bloat is another person's forward-looking design, eh? :twisted:

[Jonathan1]

Overkill on so many levels. Can Windows CE even take advantage of SMP?!?!

Dude give me a power efficient CPU and more battery life. Enough with the overkill on the bells and whistles. I like them as much as the next person but for me my #1 priority is battery life. I don't need to do 3D graphics design software on my PDA. I don't need Photoshop on my PDA. I don't need to run Unreal Tournament 2003 on the thing and I don't need a bloated OS. Why do I say that? Because invariable every time Intel comes out with a newer faster CPU Microsoft finds a way to eat as much of the hoursepower as possible for lunch. IMHO, It would be refreshing for once to see the industry stop the insanity, settle down, and give us a system that can get 20 hours off of one charge.

[Craig Horlacher]

I think having multiple core's on a pda would make a lot of sense. One can be used for smooth streaming of communications data - say, video conferancing while another keeps the user interface responsive. Using one for media playback is also a great idea...esplecially with heavy codecs like mpeg4 becoming more common.

Moblie phone pda's or pda's doing voip. I'd love to see a Vonage client that I could run on my pda to use my calling plan over wifi at a coffeehouse:)

It would be neet to see a comparison of power consumption and likely performance gain between multi core and high speed cpu's. Which whould give better batter life on average? Which would give better performance on average? Would one design win in both catagories - battery life and perfomance?

[szamot]

humbly, world domination, what else.

[kosmicki]

Can Windows CE even take advantage of SMP?!?!

I'd say not yet. But they can't handle many things yet, just wait for the next version.

Personally I can't want till PPCs have a full blown USB port. Plug any USB device in I want have have it work. (Don't start ranting about drivers, I know full and well about em ;) )

[danmanmayer]

I want a device that could do video editing on a pocket pc and then after a trip on the flight home while all the ideas are fresh inn my head i could start to edit it together.

I agree with games and compression.

Also the speech recognition will be a very processor intesive taskw hen we get that working well. I would love to dictate to word on my pocket pc.

Finally algorythmic dynamic music creation. And endless stream of new music that is similiar to all of the music you like.

[Paula]

Hi,
Can someone please explain what a multi-core processor :?: is and why it is such a good thing to have.

Paula

[Kati Compton]

Hi,
Can someone please explain what a multi-core processor :?: is and why it is such a good thing to have.
Multiple processors on a single chip. Now that we can make things smaller and smaller, multiprocessors don't necessarily have to be a separate chip for each processor.

By having things on-chip instead of on-board, you can have shorter connections between processors and more of them, decreasing the overhead of the processors "talking" to one another. Traditionally, the time spent communicating between processors in a multi-processor system has been a significant percentage of execution time - in other words, having 2x as many processors doesn't make it twice as fast.

Total performance = 2*&lt;single processor perf> - &lt;communication overhead>.

The more you can reduce the communication overhead, the closer you can get to having the performance scale better with the number of processors.

Of course, I'm skipping a lot of the details here. For example, if you are running two completely independent programs on a two-processor system, you can get the 2x speedup. But if there needs to be interaction between the two processors - that's where the problem is.

[bjornkeizers]

Hi,
Can someone please explain what a multi-core processor :?: is and why it is such a good thing to have.
Multiple processors on a single chip. Now that we can make things smaller and smaller, multiprocessors don't necessarily have to be a separate chip for each processor.

By having things on-chip instead of on-board, you can have shorter connections between processors and more of them, decreasing the overhead of the processors "talking" to one another. Traditionally, the time spent communicating between processors in a multi-processor system has been a significant percentage of execution time - in other words, having 2x as many processors doesn't make it twice as fast.

Total performance = 2*&lt;single processor perf> - &lt;communication overhead>.

The more you can reduce the communication overhead, the closer you can get to having the performance scale better with the number of processors.

Of course, I'm skipping a lot of the details here. For example, if you are running two completely independent programs on a two-processor system, you can get the 2x speedup. But if there needs to be interaction between the two processors - that's where the problem is.

Plus, you know, it looks totally sexy in marketing.

[Jonathon Watkins]

Hi,
Can someone please explain what a multi-core processor :?: is and why it is such a good thing to have.
Total performance = 2*&lt;single processor perf> - &lt;communication overhead>.



Plus, you know, it looks totally sexy in marketing.

:lol: 2 = always better than 1, right! :wink: