Hyperthreaded Desktop CPU Delivers Mixed Results

[Jason Dunn]

<a href="http://www.pcworld.com/news/article/0,aid,106898,tk,dn111402X,00.asp">http://www.pcworld.com/news/article/0,aid,106898,tk,dn111402X,00.asp</a><br /><br />Gosh - and here I was hoping that hyperthreading would lure me into buying a faster Intel CPU...but I can't say that I'm really surprised. Intel has a habit of releasing a new CPU and having the performance truly suck. After further refinement, the performance will come. Shades of Xscale anyone? The problem with that is that while I can swap out a P4 on my desktop, you can't do that on your Pocket PC.<br /><br />"Tests on three of the first systems to use the new chip indicate that hyperthreading's benefits are largely application- or even task-specific. For example, on the office applications most people use, such as Microsoft's Word and Excel, hyperthreading's impact was neutral or slightly negative. Even on specialized tests designed to highlight the benefits of hyperthreading, most improvements are negligible. Exceptions are select filters in Photoshop, some aspects of video and photo programs like Premiere and VideoWave, and certain multitasking tests.<br /><br />PC World's results also show that a PC equipped with AMD's Athlon XP 2800+ CPU, which runs at 2.25 GHz, more than held its own against these 3-GHz systems running 800 MHz faster. Hyperthreading is still new, however, and Intel says you should see improvements as developers modify their applications to take better advantage of the technology."

[Rirath]

Putting the whole hyperthreading thing aside for a moment, it's no surprise the 2800+ can hold it's own. AMD has been saying for years now it's not just the clock speed that matters. I love AMD, they're cheap and they're fast. If only they made Pocket PC processors.

[ThomasC22]

You can usually tell how useless a technology is by how much Intel hypes it...

As for this article, welllllllll, as someone who used to spend a lot of time comparing this sort of thing I can say that you can get a processor test to say whatever it wants if you don't list your methodology. Especially when your devising your own "new tests". This article was bunk.

Not that I think the result is necessarily incorrect, I don't see multi-threading causing that much of a speed jump right now, I just tend to get offended when I can tell an article is skewing results.

[Jason Dunn]

Back in the sub-500 mhz Celeron days, I had a dual 466 mhz Celeron rig on a BP6, and man, that thing smoked! Dual processors are the way to go, no question about it. It drives me :!:BONKERS:!: that more mainstream computers aren't offered in dual-proc models. Paired with an OS like Windows XP, the overall speed boost would be quite tangible. And not in raw benchmarks - overall system performance would just be much snappier, and all for a couple of hundred bucks for an extra CPU and mobo to support it.

[Foo Fighter]

It drives me :!:BONKERS:!: that more mainstream computers aren't offered in dual-proc models.

Yeah, I know what you mean. My current system is a "willamette" P4 1.7GHZ Dell 8100. I'm in the process of upgrading the RAM to 768mb. It gets the job done, but next time around I'm probably going to build my own box. A nice dually loaded to the hilt.

In hindsight, I should have waited to purchase a "Northwood" system instead. :?

[Kati Compton]

Unfortunately, I'm not all that clear on how hyperthreading actually works. Is it a superscalar processor that gets divided between the tasks? It's not *actually* dual processor... I should probably spend less time working one my own particular project, and more time reading about current processor technology. :(

[Foo Fighter]

Is it a superscalar processor that gets divided between the tasks? It's not *actually* dual processor...

Basically, it's two processor cores on one board, instead of dual processors.

[ThomasC22]

Back in the sub-500 mhz Celeron days, I had a dual 466 mhz Celeron rig on a BP6, and man, that thing smoked! Dual processors are the way to go, no question about it. It drives me :!:BONKERS:!: that more mainstream computers aren't offered in dual-proc models. Paired with an OS like Windows XP, the overall speed boost would be quite tangible. And not in raw benchmarks - overall system performance would just be much snappier, and all for a couple of hundred bucks for an extra CPU and mobo to support it.

Actually, it's funny you would make that comment after the article praised AMD since AMD is famous for almost completely ignoring the dual processor market more often than not. I still think that their attitude regarding that hurt the Athlon (e.g. it was powerful enough to be a server processor but no one would use it).

As for the reason you don't see it more in the mainstream It think it's two fold:

1 - It would cause consumer confusion because many of your average consumers just wouldn't get the concept (of a CPU much less two of them)

and

2 - Dual processors have always been Intel's little secret. Take back in the Celeron 500mhz days, Cel. 500's were $178 a piece where a P3 1Ghz was $600 and back in those days everyone was figuring out that there wasn't that big a performance advantage to the P3. I think that is why you didn't see it as an option (if you recall you could only use dual processor celerons on Abit boards because they had a lock on them).

[Kati Compton]

Is it a superscalar processor that gets divided between the tasks? It's not *actually* dual processor...

Basically, it's two processor cores on one board, instead of dual processors.

I can see how that could help running two completely separate tasks, then, which I see as the main benefit of a dual-proc machine.

I just get bad flashbacks to parallel programming class when I think about trying to get a single program to run multi-processor. In our assignments the general consensus was that the parallel version generally ran *slower* than the serial version due to communications overhead. I think it was only partly inexperience on the part of the programmers, and mostly the projects didn't have enough inherent parallelism or large enough data sets to make the overhead worthwhile...

[Stake]

Take a look at a video made between (albiet) two Intel chips; the 3.06GHz with HT on and the 3.6GHz without HyperThread enabled.

http://www.tomshardware.com/cpu/02q4/021114/p4_306ht-22.html

It's quite amazing to see the speed differences of HT especially in mulit-tasking. As stated in the article, single task tests are no longer a real measure of how well a processor performs. No one works on one program one time these days.

[ThomasC22]

It's quite amazing to see the speed differences of HT especially in mulit-tasking. As stated in the article, single task tests are no longer a real measure of how well a processor performs. No one works on one program one time these days.

The problem will still always be that not all apps are multi-threaded and explaining to a consumer that some of their apps will be faster while others will stay the same never really goes all that well.

Not that I don't applaud the effort on Intel's part none the less.

[Jonathon Watkins]

Is it a superscalar processor that gets divided between the tasks? It's not *actually* dual processor...

Basically, it's two processor cores on one board, instead of dual processors.

Well - not really. IBM has True dual processors core on one chip die - this is rather differnt. Basically a small part of the CPU is duplicated - it's not two whole cores. For a full overview see the Tech-Report on : http://www.tech-report.com/reviews/2002q4/pentium4-3.06/index.x?pg=1, but here's a snippet:

The logic needed to make Hyper-Threading work adds only 5% to the Pentium 4's die size, including duplicate copies of key resources necessary for maintaining two architectural states on one chip. Intel points out that's not much extra real estate for an enhancement that can improve performance by as much as 30% in the right scenarios.

Hyper-Threading adds so little to the Pentium 4's die size because it only requires physical duplicates of a small subset of the processor's resources. Many other CPU resources, including the caches, registers, execution units, and scheduling queue, are shared, either through static partitioning (splitting 'em in two) or dynamic sharing. The most important shared resources are the processor's execution units, where integer math, floating-point math, and load/store functions are handled. Execution stages in the deeply pipelined Pentium 4 are likely to be unused during some CPU cycles, and Hyper-Threading is intended to help keep the chip's execution pipelines busier by exposing a second logical processor.

[Kati Compton]

Well - not really. IBM has True dual processors core on one chip die - this is rather differnt. Basically a small part of the CPU is duplicated - it's not two whole cores. For a full overview see the Tech-Report on : http://www.tech-report.com/reviews/2002q4/pentium4-3.06/index.x?pg=1,

Ah. That makes more sense then. There's a link on the page there that goes to a more specific description. So it's essentially faking two processors on basically enough hardware for only one. The extra hardware is only what's needed to provide the illusion, through more complex scheduling and issue-ing.

"Only" a 5% difference in die size actually sounds significant to me, though, given how big they already are. I mean, 5% of $10 isn't much, but 5% of $1M is a lot... ;) But I suppose that's just being nitpicky about phrasing. I'd be curious as to how the change in size actually translates to real cost. IE, can they now fit fewer dies on a wafer because of that difference, or is it just that less of the space on the wafter goes unused?

[Foo Fighter]

Hmm...maybe I'm thinking of something else. I distinctly remember reading about cheap makers building twin cores on one wafer.

[Underwater Mike]

Putting the whole hyperthreading thing aside for a moment, it's no surprise the 2800+ can hold it's own. AMD has been saying for years now it's not just the clock speed that matters. I love AMD, they're cheap and they're fast. If only they made Pocket PC processors.

Yeah, if only AMD would get those faster processers in the channel! Seems like the 2400s are just now becoming widely available.

I'm waiting for nVidia to get is stuff together and for Shuttle to release the SN40. Then I'll ditch my Compaq notebook and use only my PPC for presentations.

[Will T Smith]

Hyperthreading is kinda like Dual procs "on the cheap".

The extra silicon involved with adding two slots and paying for two separate CPUs have largely thwarted multi-processor in the consumer space.

On the other hand, we now have CPUs that are capable of running 10 or more instructions per cycle. The problem here is that these capabilities are often not exploited. Especially with the P4's exceptionally long pipelines.

Hyperthreading effectively allows you to use those "unused" capabilities to 2 threads simultaneously for the same process.

Those expecting 2x results are and should be dissapointed. This wasn't the point. This really isn't a multi-processor setup (like IBMs new two CPU on one core stuff). It's simply a way to use untapped CPU resources for multi-threaded applications.

[klinux]

. It drives me :!:BONKERS:!: that more mainstream computers aren't offered in dual-proc models.

I agree. Well, all PowerMacs are dual processors which I think is a step in the right direction. But then you said mainstream.... :lol:

[ThomasC22]

Hyperthreading is kinda like Dual procs "on the cheap".


An ironic comment considering...

http://www.solutions4sure.com/product.asp?productid=1381384&affid=873&adid=873

But I know what you meant

[Jonathon Watkins]

Hmm...maybe I'm thinking of something else. I distinctly remember reading about cheap makers building twin cores on one wafer.

Yes - there has been a fair bit of talk about this tech recently. IBM are ahead of the pack here with (I believe) some of the newer PowerPC processors having this feature. Sun were seriously rumoured to be developing this functionality and Intel has been whispered about.

IBM talked about moving rapidly from one core to the other to cut down the power output – but this rather defeats the purpose of this technology if you think about it. It will probably be a while till we plebs see it on the desktop.

[Jonathon Watkins]

... all PowerMacs are dual processors which I think is a step in the right direction.

Ummm - NO. Where did you get that idea from? Some Macs do come with dual processors. Most do not.

[Janak Parekh]

... all PowerMacs are dual processors which I think is a step in the right direction.

Ummm - NO. Where did you get that idea from? Some Macs do come with dual processors. Most do not.
I think the marketing of the AltiVec unit often leads people to think it's "a second processor". While the AltiVec unit is largely separated from the main G4 core, it's not really capable of executing instructions on its own.

--bdj

[Jonathon Watkins]

I think the marketing of the AltiVec unit often leads people to think it's "a second processor".

:roll: :roll: Typical Mac mis-information. :roll: :roll: I did not realise that Mac were pushing AltiVec as a second processor. Lets face it - Mac needs all the dual processor machines they can get to be half as good as an Althlon or P4. :twisted:

[Janak Parekh]

:roll: :roll: Typical Mac mis-information. :roll: :roll: I did not realise that Mac were pushing AltiVec as a second processor. Lets face it - Mac needs all the dual processor machines they can get to be half as good as an Althlon or P4. :twisted:
Well... I surfed to Apple's site, and, lo and behold...

http://www.apple.com/powermac/

No kidding - they are offering dual-procs on every G4 desktop. 8O Mea culpa.

Mind you, G4's aren't terrible processors - they're especially great for notebooks, being low-power and low-heat - they just don't scale up as well as a core architecture. The new IBM-based Power technology will be interesting to watch.

--bdj

[Jonathon Watkins]

OK - fair enough - but the statement all PowerMacs are dual processors
is not true. If he had said "All NEW PowerMacs only come with dual processors" it would have been true. Anyway - we were talking about two processors on one core - this are just 'vanilla' dual processors. :wink:

Check this Apple quote: Twin engines. All systems have dual PowerPC G4 processors — up to 1.25GHz

Yes - an even truer statement would have read: "All NEW PowerMacs NEED to come with dual processors because even with dual 1.25Ghz processors they still get their buts whupped by Intel and AMD". :twisted: