3.2GHz Dual-Core Extreme Edition Reviews

[lithium726]

It's what I've seen! XP Home is for Single Processors but can support HTT and two threads. It wouldn't be hard for it to support Dual Cores. It will probaly support Athlon64 Dual core and the P-D 840, but not the HTT version. Doesn't make sense.

Donnie

yes, it does make sense. as long as the chip is one chip, you can put however many cores you like on it, its still one processor, and XP home will support it.

http://www.microsoft.com/licensing/highlights/multicore.mspx

Microsoft announced on Tuesday that the company won't consider dual-core, four-core, eight-core or whatever-core as individual processors, but rather that such technology will be treated, from a licensing perspective, as one processor, no matter how many cores you carve into a chip.

http://www.eweek.com/article2/0,1759,1680142,00.asp

 

[64bit_is_here]

You keep saying that, but you can't back it up. We've been through that before.
Stop saying it when there have been objections that you could not counter.

Well thats true, while you keep saying that they should both see a similar increase in performance, which you yourself did not back up.

see here , here , here and here

So the single core gets more of a performance increase with HT than Dual core, in general. Of course, some software might perhaps see a similar increase with both single and dual core, but all of the benchmarks i have seen, the single core had more of a performance increase.

Enough of this discussion. Move on

 

[Scali]

Well thats true, while you keep saying that they should both see a similar increase in performance, which you yourself did not back up.

No, I just said that in theory HT does the the same on n cores as it does on 1 core, because it is a per-core technology. And I also said that the proper software to measure these things isn't available yet, so stop posting irrelevant benchmarks... The only thing we've seen so far is that Cinebench on two HTT Xeons gives a huge boost over a single one. In fact, a single Xeon cannot beat a single core of the new Althon64 dual core, yet two Xeons with HTT beat Athlon64 with two cores. So it rather looks like the boost is only larger, not smaller, if you are using software that is properly designed for multithreading.

 

[Donnie27]

yes, it does make sense. as long as the chip is one chip, you can put however many cores you like on it, its still one processor, and XP home will support it.

http://www.microsoft.com/licensing/highlights/multicore.mspx


http://www.eweek.com/article2/0,1759,1680142,00.asp

I almost linked you to the same page.http://www.microsoft.com/licensing/highlights/multicore.mspx

As a result, Microsoft has further defined our licensing policy with regards to how server software is licensed on a system with dual-core and multicore processors.

Licensing Microsoft Software on Multicore Processors
On October 19, 2004, Microsoft announced that its server software that is currently licensed on a per-processor model will continue to be licensed on a per-processor, and not on a per-core, model. This policy will allow customers to recognize more performance and power from Microsoft software on a multicore processor system without incurring additional software licensing fees.

Last one is real good!

Q. How does this licensing policy affect products such as Microsoft Windows XP Professional?

A. Windows XP Professional is not affected by this policy as it is licensed per installation and not per processor. Windows XP Professional supports up to two processors regardless of the number of cores on the processor.

Now, as your paste of my comments show, I said "WinXP HOME", NOT WinXP Pro. I already pointed out WinXp Pro supports Dual Xeons with HT enabled, that's old news. I said WinXp Home was limited to Two threads and NOT two processors. Hyperhtreading was the only thing I saw it able to RUN. I don't know if MS updated it to run Dual Core, but I know damned will it will not run Dual Core and Hyperthreading to process 4 threads. Now if that changed, I'd like to see it, the link you gave are for MySLQ, WinXP Pro and etc., NOT WinXP Home. See here ?

Just before the credits at the bottom of the page.

For dual–processor computers, you'll need to purchase Windows XP Professional instead of Home Edition to enable the second processor to work.

She is one of the writers of Windows for Dummies.

Donnie27

 

[64bit_is_here]

No, I just said that in theory HT does the the same on n cores as it does on 1 core, because it is a per-core technology. And I also said that the proper software to measure these things isn't available yet, so stop posting irrelevant benchmarks... The only thing we've seen so far is that Cinebench on two HTT Xeons gives a huge boost over a single one. In fact, a single Xeon cannot beat a single core of the new Althon64 dual core, yet two Xeons with HTT beat Athlon64 with two cores. So it rather looks like the boost is only larger, not smaller, if you are using software that is properly designed for multithreading.

OK, Agreed.

softwares designed properly will give huge boosts. But thats software specifcally designed, and general multitasking, etc.etc will behave in the same way as the benchmarks i linked to.

Lets just wait till the introduction of the A64 Dual Core, then compare. I just hope they get it out soon

 

[lithium726]

Q. How does this licensing policy affect products such as Microsoft Windows XP Professional?

A. Windows XP Professional is not affected by this policy as it is licensed per installation and not per processor. Windows XP Professional supports up to two processors regardless of the number of cores on the processor.

that sounds like its proving what i said, XP Pro isnt being affected - the same way XP home isnt. notice they said PROCESSORS, not CORES. in this way, XP Pro could support dual 16 cored processors.

For dual–processor computers, you'll need to purchase Windows XP Professional instead of Home Edition to enable the second processor to work.

yet again, teh word PROCESSOR not CORE is used. i know that if you want to use dual processors, you have to get Pro. we arent talking about dual processors. we're talking about dual cored processors. until you give me a link that that states "windows XP Home will NOT support dual core processors with hyperthreading technology to teh full extent" i am going to take everything you say with a grain of salt - as everything else i have read - including the things you have just linked me too (or quoted) say otherwise. i would also like a link to this "XP Home cannot support more than two threads" claim.

 

[Scali]

softwares designed properly will give huge boosts. But thats software specifcally designed, and general multitasking, etc.etc will behave in the same way as the benchmarks i linked to.

Depends on what you mean by 'general multitasking'... If it's just switching between programs, and assuming that these programs are generally not CPU-intensive, well then obviously single-core with HT or dual-core without HT would already be plenty. But that's not the point, is it? Then you're again talking about responsiveness, which is not the same as performance.

 

[MonkeyShave]

No, I just said that in theory HT does the the same on n cores as it does on 1 core, because it is a per-core technology. And I also said that the proper software to measure these things isn't available yet, so stop posting irrelevant benchmarks... The only thing we've seen so far is that Cinebench on two HTT Xeons gives a huge boost over a single one. In fact, a single Xeon cannot beat a single core of the new Althon64 dual core, yet two Xeons with HTT beat Athlon64 with two cores. So it rather looks like the boost is only larger, not smaller, if you are using software that is properly designed for multithreading.

1. The proper software to test these things does exist, multiple cpu systems are not new, lots of well-written multithreaded software exists. Lots of theory on throughput, system balance, memory contention, etc... exists. Intel's multicore implementation is very similar to it's multi-cpu implementation. WRT the Cinebench list, that only includes Xeon numbers for 1 Xeon w/o HT and 2 Xeons w HT, it would be more interesting to have the other 2 configurations listed.
2. HT will provide a smaller benefit on a 2-core (whether they share a cpu package or not) system than on a 1-core system. FSB traffic (for main memory and cache coherency) becomes a larger bottleneck as the number of concurrently-executing processes increases. For example, review these benchmarks (http://www.2cpu.com/articles/42_4.html) between a dual-Xeon system and single-P4. Note how the Xeon system almost always enjoys a smaller % increase in performance due to HT than the P4 system.

 

[Donnie27]

that sounds like its proving what i said, XP Pro isnt being affected - the same way XP home isnt. notice they said PROCESSORS, not CORES. in this way, XP Pro could support dual 16 cored processors.


yet again, teh word PROCESSOR not CORE is used. i know that if you want to use dual processors, you have to get Pro. we arent talking about dual processors. we're talking about dual cored processors. until you give me a link that that states "windows XP Home will NOT support dual core processors with hyperthreading technology to teh full extent" i am going to take everything you say with a grain of salt - as everything else i have read - including the things you have just linked me too (or quoted) say otherwise. i would also like a link to this "XP Home cannot support more than two threads" claim.

Because XP Home only makes use of ONE LOGICAL CORE=P I keep saying processor because Dual Core Procs and Dual Processors are seen as the same thing last time I checked. AMD and Intel's Dual Core processors will need the same support as Dual Processors=P Did you see any site using WinXP Home for their test OS?

That's why I said it didn't make sense. Each of these cores have Local APIC's that they and the OS depend on.

Hyperthreading was built into WinXP Home, not added later. Windows XP Home would have to think that the Second core on the AMD DC Proc is an Hyperthreading Virtual Core=P See if you can get Kyle to try to setup the P-DXE 840 with WinXP Home and have it show up as 4 processors? Hell, I wish you were right on this one? Those links show MS talking about Server Software, NOT Win XP Home. This MIGHT be the same reason there's no Hyperthreading on some 8xx models as well. The second Core and HT unit would be wasted with WXP-H I suspect.

Donnie27

 

[lithium726]

yeah, but i do remeber microsoft saying all software licenseing would be on the per-processor basis, not per-core... so in theory, Home should support it... we just dont have any evidence to prove or disprove this becuase no review sites use XP home for the sake that, well, its the Home edition

 

[MonkeyShave]

See here ?

Just before the credits at the bottom of the page.

For dual–processor computers, you'll need to purchase Windows XP Professional instead of Home Edition to enable the second processor to work.

She is one of the writers of Windows for Dummies.

Donnie27

Your article dates from November 26, 2001, over 3 years before the first dual-core cpu shipped and prior to the first HyperThreading systems, you can't consider it relevent anymore. People speculated that XP Home wouldn't run on HT systems, but MSFT clarified that later. It appears that MSFT hasn't made a formal announcement about XP Home and dual-core, but I suspect that XP Home will support dual-core systems.

Hyperthreading was built into WinXP Home, not added later.

Your assertion is incorrect, XP Home shipped in 2001, the first HT systems didn't appear until 2002.

Because XP Home only makes use of ONE LOGICAL CORE=P I keep saying processor because Dual Core Procs and Dual Processors are seen as the same thing last time I checked

In that case, you have not checked recently. The CPU-Z screenshot from hardocp's P4EE 840 review (here) refers to "CPU #1 (Core 1 Logical Unit), CPU #1 (Core 1), CPU #1 (Core 2 Logical Unit), CPU #1 (Core 2)". On a 2 CPU system this displays differently. I would expect Intel's 2-Core implementation to perform similarly to a 2 cpu system, but the bios reports them differently.

 

[Lazn_Work]

The only reason XP Home only supports 1 physical CPU (2 logical CPU's) is because of licensing. There is no technical reason beyond Microsoft disabling the support in Home for it not to support as many CPU's as XP Pro they are based on the same code base. (they use the same service packs, drivers etc.)

And Microsoft has announced that they will not base its per-processor software licensing charges on the number of cores in a chip, sticking to the traditional price per processor, regardless of its number of cores. (paraphrased from Microsoft announcment)

""Microsoft software that is currently licensed on a per-processor model will continue to be licensed per processor, not per core, for hardware that contains dual-core and multicore processors," the company said in a statement." http://www.technewsworld.com/story/news/37432.html

So I would not be suprized at all if XP Home supported dual core hyperthreaded (4 logical) processors just fine. (or at min with an update)

edit: Though Microsoft's page on it is not clear, it doesn't mention XP Home: http://www.microsoft.com/licensing/highlights/multicore.mspx

==>Lazn

 

[USMC2Hard4U]

I would just like to throw out that who uses Home anyway? ewww
XP Pro for life.... well right now anyway.


And to thow it out their too, I guess that Intel is releasing the Extreme Edition and 955X Chipset late next week. (its on the front page)

w00t

 

[Scali]

1. The proper software to test these things does exist, multiple cpu systems are not new, lots of well-written multithreaded software exists. Lots of theory on throughput, system balance, memory contention, etc... exists. Intel's multicore implementation is very similar to it's multi-cpu implementation.

Not really. SMT actually IS quite new... it hasn't been employed on a large scale until Intel and IBM started offering it on their chips, not too long ago. And especially multicore in combination with SMT is quite new. Heck, before Windows XP, there wasn't even an OS that actually bothered to schedule specifically for SMT, which apparently can make quite a difference, just compare to Windows 2000. Who knows, the scheduler might need an update for dual core HT.
And most software for multicore systems is aimed at server systems, which are either not x86, or are not running Windows, or performing tasks that the average desktop user would never do (massively parallel web/fileservers, databases etc?), so they are pretty much out of scope for these particular processors.
So indeed we haven't really seen many benchmarks yet that actually tell us how good these dual core systems really are, on the desktop, with or without HT.

WRT the Cinebench list, that only includes Xeon numbers for 1 Xeon w/o HT and 2 Xeons w HT, it would be more interesting to have the other 2 configurations listed.

Yes well, just another one of those irrelevant benchmarks again. That's why I was a bit reserved and said "it looks like"...

2. HT will provide a smaller benefit on a 2-core (whether they share a cpu package or not) system than on a 1-core system. FSB traffic (for main memory and cache coherency) becomes a larger bottleneck as the number of concurrently-executing processes increases. For example, review these benchmarks (http://www.2cpu.com/articles/42_4.html) between a dual-Xeon system and single-P4. Note how the Xeon system almost always enjoys a smaller % increase in performance due to HT than the P4 system.

But then you are linking HT performance with FSB performance. That's not really fair... The same would hold for a true 4-core system, given the same amount of bandwidth.
HT itself doesn't inherently give a smaller increase in performance... You just see the standard issue of parallel computing... As you add more processing units, the overhead of distributing the workload etc etc becomes relatively larger, and eventually you'll reach a point where adding more cores (in this case logical ones) will actually decrease performance. So what we would see with a true 4-core system would also be that it would not give as much of a boost. Relative to that, the HT boost should still be about the same.
So HT in itself is not going to be any less efficient. You're just running into the standard multiprocessing problems. And how much you are affected by that, depends on the software you're using. Obviously HT can only boost performance by as much as the rest of the system will allow it (what's next... if FSB would be doubled, you'd bring up that now the HDD will become a bottleneck or something? Something will always have to be the bottleneck). My point is that if you're not running into any other problems that will degrade performance in general (as in not just with HT), you will get the same performance boost from HT as always.
Obviously HT is not directly linked to the FSB speed, so no doubt we'll see the FSB increase again in the future, which will make HT be more efficient even in the problem cases.

 

[MonkeyShave]

Not really. SMT actually IS quite new... it hasn't been employed on a large scale until Intel and IBM started offering it on their chips, not too long ago. And especially multicore in combination with SMT is quite new. Heck, before Windows XP, there wasn't even an OS that actually bothered to schedule specifically for SMT, which apparently can make quite a difference, just compare to Windows 2000. Who knows, the scheduler might need an update for dual core HT.
And most software for multicore systems is aimed at server systems, which are either not x86, or are not running Windows, or performing tasks that the average desktop user would never do (massively parallel web/fileservers, databases etc?), so they are pretty much out of scope for these particular processors.
So indeed we haven't really seen many benchmarks yet that actually tell us how good these dual core systems really are, on the desktop, with or without HT.

There exists no reason to believe that they will perform significantly differently than a dual-Xeon system (2 physical cpus, 2 cores total, 4 logical cpus total) with equivalent cache and bandwidth. Given that Intel's dual-core implementation uses separate caches, the inter-core communications overhead should be identical.

In other words, multitasking will improve for desktop apps, you can fold or encode media on your second core without slowing other apps down, and overall media encoding throughput will be faster. On the other hand, your games won't run faster (unless you're running something cpu-intensive in the background), and most other desktop apps will show little or no improvement. When you look at the benchmarks at [H], anandtech, etc... you will see exactly that.

In a few years, perhaps software vendors will update their desktop apps to take advantage of multiple cores, but due to the cost and complexity involved, but that won't happen overnight.

Yes well, just another one of those irrelevant benchmarks again. That's why I was a bit reserved and said "it looks like"...

Your earlier post misinterpreted that benchmark and implied that dual-Xeons beat Dual-Core K8s, while the single-Xeon did not beat the single K8. You interpreted it incorrectly, as the dual-Xeons used HT, while the single one did not.

But then you are linking HT performance with FSB performance. That's not really fair... The same would hold for a true 4-core system, given the same amount of bandwidth. HT itself doesn't inherently give a smaller increase in performance...

I never implied that FSB does not limit total throughput in multiple-cpu systems, everybody knows that. I stated that if enabling HT on a single-core system increased performance by, for example, 20%, then enabling HT on a dual-cpu or dual-core system will increase it by a smaller amound, probably 10-15%. In some synthetic benchmarks, you will see an identical improvement but regular applications will not experience this.
You had implied otherwise in the post I responded to.

The particular trouble with HT in this respect is that it provides a smaller benefit (at a lower $ cost) than a complete second core, these smaller returns will diminsh more quickly as you add virtual-cpus (a complete analysis would consider the effects, good and bad, of shared cache in HT systems, though empirical measurements confirm my hypothesis).

You just see the standard issue of parallel computing... As you add more processing units, the overhead of distributing the workload etc etc becomes relatively larger, and eventually you'll reach a point where adding more cores (in this case logical ones) will actually decrease performance.

I refer to exactly the standard issues of parallell computing (with regards to cacheline locking, not scheduling). HT appears to double the number of cpus in a system, therefore the overhead grows more rapidly as you add HT cores than non-HT cores.

Overall, I still consider HT a good feature, notably because it provides a measurable benefit at negligible cost, however this benefit will become smaller as the number of real cores grows.

 

[Scali]

There exists no reason to believe that they will perform significantly differently than a dual-Xeon system (2 physical cpus, 2 cores total, 4 logical cpus total) with equivalent cache and bandwidth. Given that Intel's dual-core implementation uses separate caches, the inter-core communications overhead should be identical.

Exactly, and those Xeons are also still relatively new... Only been on the market for a few years.

In other words, multitasking will improve for desktop apps, you can fold or encode media on your second core without slowing other apps down, and overall media encoding throughput will be faster. On the other hand, your games won't run faster (unless you're running something cpu-intensive in the background), and most other desktop apps will show little or no improvement. When you look at the benchmarks at [H], anandtech, etc... you will see exactly that.

In a few years, perhaps software vendors will update their desktop apps to take advantage of multiple cores, but due to the cost and complexity involved, but that won't happen overnight.

Exactly my point, we don't have the software yet to really take advantage of these new technologies on the desktop.

Your earlier post misinterpreted that benchmark and implied that dual-Xeons beat Dual-Core K8s, while the single-Xeon did not beat the single K8. You interpreted it incorrectly, as the dual-Xeons used HT, while the single one did not.

I didn't misinterpret anything, I said "it looks like". I didn't say you can be 100% sure.
Besides, we can't be 100% sure if the single Xeon used HT or not, since it was not explicitly mentioned, if you want to be pedantic.

I never implied that FSB does not limit total throughput in multiple-cpu systems, everybody knows that. I stated that if enabling HT on a single-core system increased performance by, for example, 20%, then enabling HT on a dual-cpu or dual-core system will increase it by a smaller amound, probably 10-15%. In some synthetic benchmarks, you will see an identical improvement but regular applications will not experience this.
You had implied otherwise in the post I responded to.

Obviously I implied otherwise, because we disagree on this point. I simply cannot agree with the fact that HT will implicitly give less performance on more cores. Especially with these figure you simply pull out of your head. It depends on the software used.

The particular trouble with HT in this respect is that it provides a smaller benefit (at a lower $ cost) than a complete second core, these smaller returns will diminsh more quickly as you add virtual-cpus (a complete analysis would consider the effects, good and bad, of shared cache in HT systems, though empirical measurements confirm my hypothesis).

This is because you misinterpret what HT is supposed to be doing. This is such a common misconception about HT... it really makes me sick by now...
The point of HT is not to try and give the same performance of a second core, or to be able to run multithreaded applications at all, really. The point is to make better use of the idle units in the CPU. HT is actually better off with running different threads together, if this means you will get a more varied mix of instructions, while dual-core would be fine with identical threads. This is why you need specific software to measure the full potential of HT, and this software doesn't really exist yet... which obviously is a problem with those empirical measurements you speak of. You prove the wrong thing, namely that the software doesn't make full use of the CPU, not that the CPU's technology itself becomes less efficient.

As far as your diminished returns go... As I said before, this does not just apply to HT, this applies to adding CPUs in general, whether they're logical or physical.

I refer to exactly the standard issues of parallell computing (with regards to cacheline locking, not scheduling). HT appears to double the number of cpus in a system, therefore the overhead grows more rapidly as you add HT cores than non-HT cores.

Overall, I still consider HT a good feature, notably because it provides a measurable benefit at negligible cost, however this benefit will become smaller as the number of real cores grows.

Again, the benefit will become smaller in general. Not specifically because of HT. You should look at the benefit of HT relative to the benefit of having the n physical cores.
Obviously if you add a second core to a system, that second core will already give less performance benefit than the first one... and HT has to work in the context of that second core, so you cannot expect HT to give it the usual boost AND compensate for the lack of performance benefit of the second core itself.

Oh, and I find it rather rude that you consistently misspell my nick.

 

[64bit_is_here]

Your earlier post misinterpreted that benchmark and implied that dual-Xeons beat Dual-Core K8s, while the single-Xeon did not beat the single K8. You interpreted it incorrectly, as the dual-Xeons used HT, while the single one did not.

Well if that is the case, we are yet to see a benchmark where the dual core performance increase with HT is more than the single core increase with HT.

I still think, even with proper software, the performance increase of HT on the Dual Core will at best come close to the performance gained by a single core with HT, and in general multitasking(not really cpu intensive) the performance increase(or say responsiveness) will almost be nonexistent.

Overall, I still consider HT a good feature, notably because it provides a measurable benefit at negligible cost, however this benefit will become smaller as the number of real cores grows.

Exactly my thought, but then again we are yet to see any proof of this statement.

 

[Scali]

in general multitasking(not really cpu intensive) the performance increase(or say responsiveness) will almost be nonexistent.

Problem is that it's pretty much impossible to measure that... It's a very subjective thing... Even if there isn't more actual work done, the user will think a system is more responsive when the program he is using happens to be the active thread at the time he's using it.
And at a certain point, people won't notice a difference anymore... I mean... if a button's response time is decreased from 40 ms to 20 ms, the user may notice that as more responsive... However, if it goes from 20 ms to 10 ms... will the user notice? Probably not. So it will appear that the system isn't more responsive, while actually it is twice as responsive.
Also, responsiveness is inherently a single-threaded issue. It is impossible for a user to use more than one program at the same time... He will have to switch the input between programs in some way, which he can't do as quickly as the OS switches threads. So that means that as long as a system can switch to that one thread quickly enough, it doesn't matter if it has one logical processor or 5000. And empirical evidence seems to show that people do notice an increase in responsiveness from going from one logical CPU to two logical CPUs, although most people won't be able to tell if those are one or two physical CPUs... And more than two CPUs, logical or physical, is apparently not an improvement to responsiveness. Then again, this obviously also depends on the OS and applications being used. Windows' scheduler happens to be linked to the UI, which means that the thread belonging to the active UI will dynamically get a boost, to increase responsiveness. So in Windows, you can be sure that even with a single logical CPU, that your UI thread will be scheduled quickly. In many *nix OSes however, the UI is completely independent from the OS scheduler, which means the active application will have to wait for its turn just as long as all non-active applications. I suppose if you measure Windows vs such a *nix OS, that in Windows people will probably not be able to tell any difference between 2 logical cores and more, but the *nix OS will still feel more responsive with more than 2 logical cores.

 

[Donnie27]

yeah, but i do remeber microsoft saying all software licenseing would be on the per-processor basis, not per-core... so in theory, Home should support it... we just dont have any evidence to prove or disprove this becuase no review sites use XP home for the sake that, well, its the Home edition

True HEHEHEHE!

Donnie27

 

[Donnie27]

Your article dates from November 26, 2001, over 3 years before the first dual-core cpu shipped and prior to the first HyperThreading systems, you can't consider it relevent anymore. People speculated that XP Home wouldn't run on HT systems, but MSFT clarified that later. It appears that MSFT hasn't made a formal announcement about XP Home and dual-core, but I suspect that XP Home will support dual-core systems.

WinXP Pro shipped with Dual Xeon Workstions that used Hyperthreading long before the 3.06GHz shipped with Hyperthreading. Win XP Home shipped with Hyperthreading from day one=P My buddies at Maximum PC tested it.

Your assertion is incorrect, XP Home shipped in 2001, the first HT systems didn't appear until 2002.

See above?

Her that link was from when Intel shipped Dual Core Xeons with HTT, sorry, you're wrong. Again, the Second Core would have to look like HTT to the OS or it will not work with WinXP home=P I know WinXP Pro can run 4 threads, I said WinXP Home couldn't. Hell, I'm just glad folks are talking and not being mean about it it!

In that case, you have not checked recently. The CPU-Z screenshot from hardocp's P4EE 840 review (here) refers to "CPU #1 (Core 1 Logical Unit), CPU #1 (Core 1), CPU #1 (Core 2 Logical Unit), CPU #1 (Core 2)". On a 2 CPU system this displays differently. I would expect Intel's 2-Core implementation to perform similarly to a 2 cpu system, but the bios reports them differently.

Too bad this thread is too long for you to read all the posts. You're posting stuff I already bitched about. Dual Core so far looks just like Dual Processor, and NOT HTT. I complained that it's BOGUS for Intel to act like a Crack Dealer and get folks hooked, then charge them for something that has been almost free up to now. It has nothing to do with how well it works or not, it's screwed up before that point is reached. Because if it doesn't work as well, they're being even slimmmmmier!

Kyle and the Gang could solve this pretty quickly. I'm sure many AMD owners want to know if WinXP Home is will work like you guys are saying. Because it can work with HTT doesn't mean it can work with dual Core Processor. One more Time, I did say unless it is Patched with another Service Pack or etc.........

Donnie27

 

[MalfurionStormrage]

Well, I'd be more than happy to test if someone wants to ship me a dual-core Intel chip and a 955 motherboard! My MSDN archive library goes all the way back to 1993... Heck, I'd even be willing to check using the betas of XP Home and Pro to see when they may have slipped in HT support.

 

[LionKinG88]

nice sites.. I've always been an Intel fan

 

[Donnie27]

Well, I'd be more than happy to test if someone wants to ship me a dual-core Intel chip and a 955 motherboard! My MSDN archive library goes all the way back to 1993... Heck, I'd even be willing to check using the betas of XP Home and Pro to see when they may have slipped in HT support.

Hehehehe! Me TOO!

Donnie27

 

[Rhys]

I came across this document off Microsoft's site which some of you may or may not find interesting. http://www.microsoft.com/whdc/system/CEC/HT-Windows.mspx

XP Pro will support a maximum of 4 logical processors and Home will support 2. This would mean that even though the licensing model for Home supports 1 physical processor with any number of logical processors, it will not actually support an extreme edition with dual core and hyperthreading since that's four logical processors.

It's possible that SP2 addresses this issue or a future service pack will or the document from Microsoft is actually incorrect.

 

[lithium726]

I came across this document off Microsoft's site which some of you may or may not find interesting. http://www.microsoft.com/whdc/system/CEC/HT-Windows.mspx

XP Pro will support a maximum of 4 logical processors and Home will support 2. This would mean that even though the licensing model for Home supports 1 physical processor with any number of logical processors, it will not actually support an extreme edition with dual core and hyperthreading since that's four logical processors.

It's possible that SP2 addresses this issue or a future service pack will or the document from Microsoft is actually incorrect.

updated: May 1, 2003.

that is two years old, and only mentions HT. has nothing to do with dual core.

edit: wow, i have no idea why i put 5... i guess i was thinking about 2005 or something...

 

[Donnie27]

updated: May 1, 2003.

that is five years old, and only mentions HT. has nothing to do with dual core.

Hehehe, that's 2 years old. I hope it works like you're saying.

Donnie27

 

[Chris_Morley]

I just had to mention that I have an overclocked dual core Extreme Edition @ 4Ghz on my desk for review...

Sweet.

 

[Jonsey]

I just had to mention that I have an overclocked dual core Extreme Edition @ 4Ghz on my desk for review...

Sweet.

How's the heat output?

 

[Rhys]

Hehehe, that's 2 years old. I hope it works like you're saying.

Donnie27

I understand the article is 2 years old. XP is even older. I'd like to see something definitive that shows a Windows XP Home edition running one physical processor with four logical processors or Windows XP Pro using two physical processors with 4 logical processors each. Until I see that, I'm holding off on whether or not we will be limited to 2 logical processors for Home and 4 for Pro. I understand the licensing of the physical processors but that doesn't mean it will technically function.

I personally won't care much one way or the other if we can't get two physical processors with dual core and HT into a box anyway since this will all be moot.

 

[Donnie27]

I understand the article is 2 years old. XP is even older. I'd like to see something definitive that shows a Windows XP Home edition running one physical processor with four logical processors or Windows XP Pro using two physical processors with 4 logical processors each. Until I see that, I'm holding off on whether or not we will be limited to 2 logical processors for Home and 4 for Pro. I understand the licensing of the physical processors but that doesn't mean it will technically function.

I personally won't care much one way or the other if we can't get two physical processors with dual core and HT into a box anyway since this will all be moot.

Right! Yet I know folk who did buy WinXP home because it was cheaper and Hyperthreading worked. There are many more millions of Customers worried about Hyperthread and Dual Core than AMD's x86-64 that only make a very small fraction of total PC's on the market. Many of them don't care about WinXP 64 Pro either, who cares LOL!

I'll wait to see as well.

I was only saying that unless something changed or is fixed, current Dual Core procs look like two physical Processors and not Logical processors. WinXp only support one proc. Now, all they'd need to do adjust/tweak the Arbiture Logic on Intel Chips, or the Crossbar Interface on AMD chips to tell the OS that these are Logical, not Physical, then they could easily work with WinXP Home just as Hyperthreading does. I agree that there is nothing set in stone.

Last time I read something about this, it was said that there are many more WinXP Home users than Pro. Of course that's not the case in the segment.

Donnie27

 

[Chris_Morley]

How's the heat output?

It's watercooled, but I haven't checked heat yet, just received it last night...

 

[MonkeyShave]

Exactly, and those Xeons are also still relatively new... Only been on the market for a few years.

That doesn't matter, I found the HT-specific programming info shortly after launch, anybody can find that material now with a small amount of research.

Exactly my point, we don't have the software yet to really take advantage of these new technologies on the desktop.

Doesn't matter, even when it arrives (several years away), it won't see any larger improvements than that seen in workstation apps with easily parallalizable work. Benchmarks designed for workstations and servers will show the upper bound for how much benefit desktop apps can gain.

I didn't misinterpret anything, I said "it looks like". I didn't say you can be 100% sure.
Besides, we can't be 100% sure if the single Xeon used HT or not, since it was not explicitly mentioned, if you want to be pedantic.

Being pedantic, We do know. Referring to the benchmark you mentioned earlier (Cinebench 2003), we see 4 significant rows.

Xeon 3,6 GHz - L2 2M        [b]4 logical cpus[/b] ---  38,4
Athlon Dual Core 2.400 Mhz  [b]2 phisical cpu[/b] ---  41,4
Athlon Dual Core 2.400 Mhz  [b]1 phisical cpu[/b] ---  77,5
Xeon 3,6 GHz - L2 2M        [b]1 phisical cpu[/b] ---  83,4

Since the same chart refers to these items:

Pentium 4 EE 3,46 Ghz       [b]2 logical cpus[/b] ---  63,4
Pentium 4 EE 3,46 Ghz       [b]1 phisical cpu[/b] ---  75,1

We may deduce that the phrase X logical cpus implies HT where X phisical (their spelling) cpus implies no HT. Therefore the best Xeon score used 2 chips + HT. If the chart included the 1-chip, HT-enabled and 2-chip, HT-disabled numbers, you would see differences similar to that measured everywhere else, i.e. enabling HT on a 2 cpu system provides a smaller benefit than enabling it on a 1 cpu system.

Obviously I implied otherwise, because we disagree on this point. I simply cannot agree with the fact that HT will implicitly give less performance on more cores. Especially with these figure you simply pull out of your head. It depends on the software used.

In an earlier post, I linked to a 2cpu.com article which contains text like:
"We see an average improvement of ~52.5% when hyper-threading is enabled on the P4 and an average improvement of ~28% with the Xeons."

or

"It [HT on 2 CPU Xeon] accounted for a ~11% decrease in overall encode time. The P4C test system shows a remarkable 30% decrease in overall encode time."

In their Cinebench 2003 results, we see:
Xeon, no HT, 1 CPU: 279
Xeon, no HT, 2 CPU: 519
Xeon, with HT, 2 CPU: 591
Bringing the second CPU in (without HT) boosted speed by 86%.
Enabling HT on 2 cpus brought a 14% improvement

P4, no HT: 323
P4, with HT: 381
An 18% improvement

In almost every benchmark that they ran, the 2 CPU system demonstrated a smaller benefit from HT than the 1 CPU system. The problem isn't specific to HT, and that doesn't make HT a bad tech, but you are behaving irresponsibly when you say:

So it rather looks like the boost is only larger, not smaller, if you are using software that is properly designed for multithreading.

You prove the wrong thing, namely that the software doesn't make full use of the CPU, not that the CPU's technology itself becomes less efficient.

Refer to Amdahl's Law, which describes why many algorithms (aside from embarassingly parallell problems) cannot significantly decrease job completion time by adding processing units. While software can obviously perform worse than this theoretical maximum, it make larger gains.

Again, the benefit will become smaller in general. Not specifically because of HT. You should look at the benefit of HT relative to the benefit of having the n physical cores.
Obviously if you add a second core to a system, that second core will already give less performance benefit than the first one... and HT has to work in the context of that second core, so you cannot expect HT to give it the usual boost AND compensate for the lack of performance benefit of the second core itself.

Oh, and I find it rather rude that you consistently misspell my nick.

I had not intentionally misspelled your nick, sorry about that.

You have argued that HT remains as effective when added to a 2 cpu system than when added to a single cpu system, if this does not imply that you will see an improvement equal to what HT provides on a 1 cpu syste, then what does it mean?
You have stated that it is more effective.
In this last paragraph, you contradict yourself, saying that obviously HT has a smaller effect on the second core.

Perhaps you are taking the nonsensical position that it is the second real core that cripples HT on that core. That position makes no sense, as without the second real core, you wouldn't have a second HT core.

 

[MonkeyShave]

Her that link was from when Intel shipped Dual Core Xeons with HTT, sorry, you're wrong. Again, the Second Core would have to look like HTT to the OS or it will not work with WinXP home=P I know WinXP Pro can run 4 threads, I said WinXP Home couldn't.

Her article dates from November 26, 2001.
Lets look at some intel and Microsoft press releases.
8/24/2001, Microsoft Windows XP Release To Manufacturer
10/25/2001, Windows XP Ships
2/25/2002, Intel ships first HT Xeons

Therefore the article you linked predates Intel's first HT cpu shipment. Additionally, XP shipped prior to the first HT cpus from intel, exactly what I claimed.

You may ask "How, then, did XP support HT out of the box?" Intel and MSFT worked to specify how the bios should report HT cpus. If the bios reports a 2 core HT cpu as a single CPU (which would make a ton of sense), then, from a technical perspective, XP Home will run on a dual core HT cpu.

All of the material Microsoft has published regarding multi-core cpus indicates that they will license per Physical Processor (defined in MSFT docs as a single package, regardless or # of cores). Zero published info states that XP Home will not support a second core, with or without HT. You have made tenuous leaps of logic and made many claims that XP Home will not support a second core with HT, but you have offered no evidence of this.

If you have some reason, with evidence (other than your nebulous claims that dual core "looks like" dual processor (disregarding the different CPU-Z output) and that the bios will report a dual core system as having 2 processors (which seems rather unlikely), then you should present that evidence.

If you do not have such evidence, you should not make claims that you cannot support.

One more Time, I did say unless it is Patched with another Service Pack or etc.........

Donnie27

If MSFT provides a patch or Service Pack, then the point is moot, the important question is one of Microsoft's intentions, the particular technical capabilities of a 3.5 year old box of XP Home don't really matter (not that I've ever owned a copy of XP Home). Microsoft's public statements about multicore cpus indicate that they will treat them (from a legal and licensing perspective) identically to single-core cpus. There exists no reason to believe that they would treat XP Home differently.

Furthermore, they would need to patch XP Pro in a similar fashion, as your claims imply that it too was written without any knowledge of multicore. Microsoft has made an official announcement about XP Pro (linked previously) and stated that it will support 2 multi-core cpus (with or without HT). It would detect #cpus identically to XP Home, so both systems would need a similar patch, but I suspect that neither will need a patch for basic support.

 

[Scali]

That doesn't matter, I found the HT-specific programming info shortly after launch, anybody can find that material now with a small amount of research.

Anyone with half a brain can figure out how to use HT even without any special info...
That doesn't mean that all software magically becomes optimized for HT overnight.

Doesn't matter, even when it arrives (several years away), it won't see any larger improvements than that seen in workstation apps with easily parallalizable work. Benchmarks designed for workstations and servers will show the upper bound for how much benefit desktop apps can gain.

In that case, people should be benchmarking with workstation/server apps, which I haven't seen in any review yet.
I don't fully agree by the way... While there certainly have been many multiprocessing apps for workstations/servers over the years... I've seen none yet that are specifically optimized for HT... They all just treat a HT system as a regular multiprocessing system, which is obviously not optimal. Running the same thread twice on a physical CPU is generally not a good way to make sure that the second thread will take advantage of the idle units from the first thread.

We may deduce that the phrase X logical cpus implies HT where X phisical (their spelling) cpus implies no HT. Therefore the best Xeon score used 2 chips + HT. If the chart included the 1-chip, HT-enabled and 2-chip, HT-disabled numbers, you would see differences similar to that measured everywhere else, i.e. enabling HT on a 2 cpu system provides a smaller benefit than enabling it on a 1 cpu system.

No we don't. Two logical CPUs are still one physical CPU. We can merely guess what they imply. Not that I don't agree with you that this is probably what they meant, but if you want to be pedantic, do it the right way.
We can't know for sure, because obviously this review was again written by someone who doesn't really grasp the whole subject.

In an earlier post, I linked to a 2cpu.com article which contains text like:
"We see an average improvement of ~52.5% when hyper-threading is enabled on the P4 and an average improvement of ~28% with the Xeons."

or

"It [HT on 2 CPU Xeon] accounted for a ~11% decrease in overall encode time. The P4C test system shows a remarkable 30% decrease in overall encode time."

In their Cinebench 2003 results, we see:
Xeon, no HT, 1 CPU: 279
Xeon, no HT, 2 CPU: 519
Xeon, with HT, 2 CPU: 591
Bringing the second CPU in (without HT) boosted speed by 86%.
Enabling HT on 2 cpus brought a 14% improvement

P4, no HT: 323
P4, with HT: 381
An 18% improvement

In almost every benchmark that they ran, the 2 CPU system demonstrated a smaller benefit from HT than the 1 CPU system.

This is wrong. What is right is "Going from 2 logical processors to 4 logical processors gives a relatively smaller boost than going from 1 logical processor to 2 logical processors".
You would have to do a lot more benchmarking to isolate the effect of HT...
One test that should be done is between 2 and 4 physical processors, without HT.
I can pretty much guarantee you that you will see a similar reduction in gain, simply because the software itself becomes less efficient when having to manage more threads, and it will be come more bandwidth-starved, and things like that.

Refer to Amdahl's Law, which describes why many algorithms (aside from embarassingly parallell problems) cannot significantly decrease job completion time by adding processing units. While software can obviously perform worse than this theoretical maximum, it make larger gains.

This is exactly the kind of stuff that I mean. It's not HT, it's this kind of generic limitations of multiprocessing, that's what you're seeing.

You have argued that HT remains as effective when added to a 2 cpu system than when added to a single cpu system, if this does not imply that you will see an improvement equal to what HT provides on a 1 cpu syste, then what does it mean?

It means that HT is a core-local technology... it affects one core only. And if nothing else affects that core, HT can do exactly the same for 1 core as it can do for N cores. By comparison... the ALU is also core-local... The ALU doesn't magically get slower when you add more cores, does it? It's completely unaffected.
HT simply allows you to use the idle units in the core... No matter how many cores you have, each core has its own HT which does this. The only problem is feeding all these cores. But as I say, that's a generic problem. Whether you have logical or physical processors, you run into the same problems.

 

[Donnie27]

No it doesn't and I didn't. Microsoft said right from the start that WinXP Pro was meant for Dual Processor system with Hyperthreading or NOT=P Hyperthreading was talked about as far back as 1999 when Intel was talking about Jackson Tech and for Linux

The Geek Viewpoint

I'm saying and will keep saying that unless I see something different, Dual Core Looks like TWO PROCESSORS and WinXP Home only supports ONE PROCESSOR=P WinXP Home will not initialize the needed Local APIC for the second Processor, a Logical processor doesn’t need this. For these versions of Windows the license only applies to installs because the OS does the rest. If WinXP Home can see Core #2 as a Logical Processor, then it will work as I already said. There's NO WAY IN HELL WinXP Home is going to see 4 cores and process 4 threads in its CURRENT form. The first Foster Xeons shipped with HT disabled, then turned on the abotu 8 months later. WinXP Pro and showed 4 cores and processed 4 threads it.

Just like EV6, Integrated memory controllers, multi-RAM channels and etc.., Alpha keeps on giving us great tech.

http://techreport.com/onearticle.x/1947
http://techreport.com/columns/ryu/2001q1/smt/
http://www.extremetech.com/article2/0,1558,75740,00.asp

Donnie27

 

[lithium726]

Dual Core Looks like TWO PROCESSORS and WinXP Home only supports ONE PROCESSOR

and microsoft has said that a DUAL CORE processor will be treated as ONE PROCESSOR in its operating systems!

i have an idea. either wait until we have evidence of both claims, or make morley test it

There's NO WAY IN HELL WinXP Home is going to see 4 cores and process 4 threads in its CURRENT form.

proof? you are the only person in this conversation that is saying this.

edit: i just PMed Morley requesting that he install XP Home to test it out we'll see what he says

 

[Donnie27]

and microsoft has said that a DUAL CORE processor will be treated as ONE PROCESSOR in its operating systems!

i have an idea. either wait until we have evidence of both claims, or make morley test it

proof? you are the only person in this conversation that is saying this.

edit: i just PMed Morley requesting that he install XP Home to test it out we'll see what he says

Microsoft said "license" and WinXP Pro, not processor and WinXP Home. You're the only one saying otherwise! The link showed that.

Donnie27