P4 & hyper threading a "no-go" in Win2K?

[Ron_Jeremy 0]

Many of us already know the new 3.06GHz & higher P4's will have hyper threading enabled, similar to their Xeon brothren. But, from reading some of the columns over at HardwareCentral, it looks like HT may mean switching to WinXP. That sux Here's a snippet below;

 

Conventional wisdom would suggest that, due to their inherent multiprocessor support, Windows XP Professional and Windows 2000 would be the optimal OS choices for a Hyper-Threading desktop (Win NT is getting a bit long in the tooth). According to Intel, this is only half right -- while Win XP Pro is indeed on its list of recommended operating systems, Win 2000 is not. Oddly, Intel goes so far as to recommend that Hyper-Threading be disabled for use with Windows 2000, NT, 98, 98SE, and Me.

 

Odder still, Windows XP Home Edition is on the recommended list, Intel listing it alongside Win XP Pro as products that "include optimizations for HT Technology." We know that Windows XP Home does not officially support multiple processors, while Win 2000 does; this makes the Intel list look a tad backwards.

 

Either Intel is working some marketing magic, or Microsoft can hit a switch (Windows Update, anyone?) and flip some hidden switch in Win XP Home. For the record, Intel says it's also "working with the Linux community" to get popular distributions up to speed, though only Windows-based PCs will get to wear the nifty Intel HT Technology sticker.[/url]

[tweaked 0]

With Win2k you lose performance in many things with hyperthreading, it (win2k) can not tell the difference between 2 processors and 2 virtual processors. SMP wil always be "on" and always try to distribute the workload accordingly.

 

Supposedly, Winxp will auto toggle between using the cpu as a single proc, and using it as a dual proc depending on the application. (100% resources to a single thread, or break up multiple threads between the two virtual processors... err something like that. The description i got was a little over techno babbly.) It will be completely transparent to the user of course.

 

Plus various other OS level optimizations for virtual multiple processors.

[Jerry Atrik 0]

i have hyperthreaded p4 xeons. w2k doesnt utilize them

xp pro does pretty well, only to a point. the only thing xp pro takes advantage of is running multiple apps. occasionally jumping to 30% cpu (out of 400%)

 

only a few of the best and most expensive products really take advantage of the ht tech.

pro/e at times the cpu's will sit at 60% for 5 or 6 seconds.

[isochar 0]

Run 4 SETI clients on your rig with hyperthreading enabled and you'll see a 2:53 per WU per processor! (Using 2.0's)

[Jerry Atrik 0]

ive never played with seti

saw it once

[Admiral LSD 0]

Quote:

With Win2k you lose performance in many things with hyperthreading, it (win2k) can not tell the difference between 2 processors and 2 virtual processors. SMP wil always be "on" and always try to distribute the workload accordingly.

Supposedly, Winxp will auto toggle between using the cpu as a single proc, and using it as a dual proc depending on the application. (100% resources to a single thread, or break up multiple threads between the two virtual processors... err something like that. The description i got was a little over techno babbly.) It will be completely transparent to the user of course.

Plus various other OS level optimizations for virtual multiple processors.

So let me get this straight, unless the app being run is multithreaded the CPU stays in "Single CPU" mode so in effect, this is going to get wasted on the vast majority of people who only buy PCs for basic web surfing and email (neither of which utilise multithreaded apps)?

On another note, how does HyperThreading compare with a "real" 2 or 4 CPU system. I would imagine it'd be slower since it uses one CPU to emulate two but is that really the case and if so, by how much?

[Jerry Atrik 0]

for the most part that is true

xp, from a system standpoint, does use the virtual cpu(s) to pre-cache info for the physical one. works darn nice if u are one who has many things open and running at the same time.

 

on the flip side some of the cheapy "not very well written" programs will actually run slower on a hyperthreaded machine, others see no speed enhancement at all.

 

it's a new technology that intel introduced early this year, and it takes awhile for developers jump on the boat.

ive always found that 2 slower cpu's is faster and better than 1 super fast one

[CUViper 0]

Quote:

On another note, how does HyperThreading compare with a "real" 2 or 4 CPU system. I would imagine it'd be slower since it uses one CPU to emulate two but is that really the case and if so, by how much?

I agree that it's probably slower than a true multiprocessor system, but not as much as you would think. P4's, like most new processors, are superscalar, meaning they can run more than one instruction in a single clock cycle. Normally, this is done by finding instructions in a program that aren't dependent on each other and running them in parallel. Sometimes you can't find eligible instructions for this, and you end up running only one instruction in a clock cycle with the other pipelines wasted.

What the hyperthreading basically does is fill these unused pipelines with instructions from other threads. The "filler" threads could be from the same program (if it's multithreaded), or even from other programs, like system processes. Since threads are by nature independent of each other, they are safe to run in parallel on a superscalar processor. Of course it's not quite that easy, since different threads expect different values to be in registers, but that's the basic idea.

In most cases, the system will make better use of the processor time, so a hyperthreaded CPU will be faster than a normal single CPU. But since a dual-CPU setup still has more pipelines to play with, it will still be faster. Hyperthreading therefore should fall somewhere in-between a single and dual CPU system.

Of course I could be wrong...

BTW, does anyone know if Intel is going to make dual-processor, hyperthreading-enabled systems?

[Admiral LSD 0]

That makes sense.

 

Quote:

BTW, does anyone know if Intel is going to make dual-processor, hyperthreading-enabled systems?

 

They already do. HT was in Xeons long before it came to the P3. Jerry Atriks Xeons are HyperThreaded. Trouble is, Xeons cost more than regular P4s and require special motherboards due to a different socket design. In addition, most Xeon boards are server oriented (due to the chips main focus being the server environment) and as such, don't include things like AGP slots. A few "Workstation" boards but these are all RDRAM based I believe.

[Jerry Atrik 0]

very true

when i got this rig (april?) there were only 2 workstation boards and they are rambus only. but rambus is good and fast but just a little more expensive.

 

i never heard of supermicro before i got this board but i am happy with it

not a cheesy board at all

[tweaked 0]

from what i understand, it will behave like a single cpu when a single thread is consuming near 100% of its time, but with multithreaded apps and lower cpu utilistaion the cpu will run multiple threads, thereby speeding up multi-threaded apps a bit, and multitasking a bit.

 

something like that?

 

cr@p, been too long, don't remember quite right.

 

i can't remember if that was the OS that was controlling that or the chip... probably both to some extent. I Don' think the intel rep specified.

 

Multithreaded apps written with HT in mind will run much better on the new p4 than multithreaded apps that were really written for 2 cpus.

[Brian Frank 0]

Quote:

very true
when i got this rig (april?) there were only 2 workstation boards and they are rambus only. but rambus is good and fast but just a little more expensive.

i never heard of supermicro before i got this board but i am happy with it
not a cheesy board at all

SuperMicro is arguably the best mobo company. Them or Tyan. They cost a good chunk of change---but it's worth it.