• Some users have recently had their accounts hijacked. It seems that the now defunct EVGA forums might have compromised your password there and seems many are using the same PW here. We would suggest you UPDATE YOUR PASSWORD and TURN ON 2FA for your account here to further secure it. None of the compromised accounts had 2FA turned on.
    Once you have enabled 2FA, your account will be updated soon to show a badge, letting other members know that you use 2FA to protect your account. This should be beneficial for everyone that uses FSFT.

Windows 7 Boosts Hyper-Threading Support

HardOCP News

[H] News
Joined
Dec 31, 1969
Messages
0
InformationWeek has an article on Hyper-Threading enhancements in Windows 7 that is fairly interesting. I know most of you have forgotten about Hyper-Threading but since Intel brought it back, Microsoft ought to be taking advantage of it, right?

Windows 7's improved hyper-threading support could be good news for users of processor-intensive applications, such as 3-D games and graphics programs. The catch is that the applications themselves must be written in such a way as to take advantage of hyper-threading. To date, only a small percentage of apps include that capability.
 
I'm fine with whatever they can do to tweak out as much performance as they can out of 7.
 
Same problem with multi-cores....software has to be written to take advantage of it. AMIRITE?
 
As long as windows supports it, it means that windows own functions will be faster.. thats got to be good for something at the very least :)
 
The catch is that applications have to actually use the technology? Since when was this a catch? My bad, I forgot with a Mac the magic pixie dust would make everything take advantage of HT.
 
The catch is that applications have to actually use the technology? Since when was this a catch? My bad, I forgot with a Mac the magic pixie dust would make everything take advantage of HT.

All joking aside, I wonder how well this is going to work in OS 10.6 with Grand Central. Its supposed to be one of the major major under-the-hood tweaks there (OpenCL is the other). We'll know in a few months.
 
The article almost seems like they are confusing Hyperthreading with Multiple-core processors in general.

"Bill Veghte, Microsoft's senior VP for Windows development, said the software maker has been working closely with chipmaker Intel (NSDQ: INTC) to bolster Windows 7's ability to get the most from hyper-threading -- an architecture where processing is divvied up among multiple processors or cores."

Nevertheless, I have an older Dual Xeon box; two processors both with HT for a total of 4 "virtual" processors. The thing still hauls ass to this day, and I look forward to putting windows 7 on it.
 
OS "Awareness" of hyperthreading is crucial for good performance.

I have a 3.0GHz 8 core (16 hyperthreads) Nehalem Mac Pro... and OSX 10.5 is NOT hyper-threading aware and screws it up all the time.

I write massively parallel scientific simulation code for a living... so I get the most out of every core available to me.

The problem with an OS not being aware of hyper-threading is that it will happily schedule two compute intensive tasks to the same physical CPU (one on each of the two hyper-threads the CPU shows up as). This is NON-OPTIMAL when there are other free physical cores laying around not getting used because these two processes will compete for compute resources (registers, floating-point pipelines, L2 cache, etc.) when they could be running happily on separate cores.

Now Nehalem is not as bad in this situation as previous hyper-threaded CPUs were. It has a big, fast shared L3 cache that limits cache misses... and the upgraded memory architecture (finally catching up with AMD) means that the penalty is smaller for cache misses that do happen. BUT.... like I say... you still get performance degradation over running the processes on separate physical cores.

I don't know about Windows (can't use it... since it's not UNIX based)... but Linux does a wonderful job of scheduling processes on hyper-threaded processes. It diligently keeps them off the same physical CPUs when possible.... you can watch it move processes around as they get spawned and settle in on a single CPU.

Like I mentioned earlier OSX 10.5 is TERRIBLE. If I'm running an 8 processor job on my Mac Pro it will happily schedule all 8 processes to _4_ physical CPUs! Even worse... since this is a dual-quadcore machine.... all _8_ of those processes share L3 cache! All while a _whole_ physical CPU sits empty.

I've raised this issue to our Apple representatives and are working with them to get this corrected before Snow Leopard comes out.... let's hope it does....

One more thing.... putting two _threads_ of the same process that are working on the same data on one core can actually _improve_ performance a lot. This is because the data will be sitting in cache for both threads... and it keeps communication down a lot. This works really well in a producer / consumer threading model where one thread is setting up work packages while another is doing computations on those packages. If both of those threads are on the same hyper-threaded core you can see some ridiculous speed increases. In this regard OSX does really well. Their threading model allows you to give hints to the scheduler that two threads will be or won't be using the same data... and the scheduler will try to put them closer / further away from each-other physically.

Anyway... this is a long-winded way of saying that the OS being "aware" of hyper-threading is critical to getting good performance. Microsoft saying that they are building in support for hyper-threading is definitely a good thing...

Friedmud
 
OS "Awareness" of hyperthreading is crucial for good performance.

I have a 3.0GHz 8 core (16 hyperthreads) Nehalem Mac Pro... and OSX 10.5 is NOT hyper-threading aware and screws it up all the time.

I write massively parallel scientific simulation code for a living... so I get the most out of every core available to me.

Hate to be an ass, but do you really have a Mac Pro? I mean, there is no such thing as a 3.0GHz Nehalem Mac Pro. That model is not available. Only 2.26, 2.66, and 2.93GHz models are available. I mean, someone like you must know the speed of your own computer, shouldn't you?
 
/me waits for him to come back and say it's the 2.93 GHz model which is, for all intents and purposes that anybody would give a fucking rats ass about, 3 GHz effective speed.

Semantics... they'll kill ya...
 
Hate to be an ass, but do you really have a Mac Pro? I mean, there is no such thing as a 3.0GHz Nehalem Mac Pro. That model is not available. Only 2.26, 2.66, and 2.93GHz models are available. I mean, someone like you must know the speed of your own computer, shouldn't you?

Did you just flame him for rounding 2.93 to 3.0? Give me a break, what a toolbox.
 
Hate to be an ass, but do you really have a Mac Pro? I mean, there is no such thing as a 3.0GHz Nehalem Mac Pro. That model is not available. Only 2.26, 2.66, and 2.93GHz models are available. I mean, someone like you must know the speed of your own computer, shouldn't you?

Sigh... Yes it's 2.93Ghz.... Look if you want to get technical about it it can dynamically overclock individual cores by up to 300Mhz (it's called turbo boost... And depends upon power and thermal constraints)... So it's really a 2.93-3.23ghz Nehalem... But I think I can be forgiven for rounding a bit....

I'm glad others found the post worth reading... Even with this glaring "mistake". ;-)

Friedmud
 
Is there a reason to own a Mac Pro is a better question?

The base $2500 mac pro is a POS spec-wise; and the prices seem to go up by extremes for every little upgrade..

Well...he did mention the dual-processor thing. I am not familiar with any dual-processor mainstream motherboards...
 
Intel makes several dual CPU boards for Xeons (as regular Core 2 Duos aren't really designed for that kind of operation, and the current Xeons are Core 2 Duo architecture processors anyway). Supermicro has a few Tyan a few, and I think Gigabyte has one or two iirc. They aren't considered "mainstream" no, but at a forum like this, I'll bet you there's more than a few people running current dual Xeon boxen these days...

If I had the money to do it I'd get such a beast, with dual Xeons as fast as I could, with 32GB of RAM on it, and a lot of other goodies too. :)
 
Software taking the advantage of multiple threads and also cores is always nice. Hope this will actually start to bring some performance improvements.
 
Is there a reason to own a Mac Pro is a better question?

The base $2500 mac pro is a POS spec-wise; and the prices seem to go up by extremes for every little upgrade..

For a 4 or 8 core Xeon workstation with ECC RAM, it is actually among the cheapest out there. Ever since Apple went from G5s to Xeons for the Mac Pros, they have been cheaper than same-specced machines by Dell or HP by $500 to $1000. The problem is that there is a very high base minimum spec with the Mac Pro, while with something like a Dell Precision you can tweak the specs much much lower. The other problem is that most people don't need such a machine. I certainly don't, even though I run Final Cut Studio on my Macs. A Core 2 Duo or consumer level i7 is plenty for me, and IMHO a desktop Mac with that level of CPU with PCI-E expansion is the only real hole in Apple's hardware lineup.

If you need a powerhouse workstation though, its well worth it, otherwise I suspect Apple, Dell, HP, etc, wouldn't bother selling multi-CPU Xeons in the first place.
 
If I had the money to do it I'd get such a beast, with dual Xeons as fast as I could, with 32GB of RAM on it, and a lot of other goodies too. :)

Throw an Intel SSD in there and I'm pretty sure the thing is fast enough to send you back in time.
 
Well, still wont change my view on Core I7 for now.

A masterpiece who doesnt really stand out as the product "must have" while a tricore from amd can almost do it, and i guess a dualcore nahelem could do it better than the tricore!.

I wonder if intel is gonna launch a dualcore I5 with HT cause that will be a killer chip!
 
Throw an Intel SSD in there and I'm pretty sure the thing is fast enough to send you back in time.

Bleh, that would just slow it down... was planning to create a 24GB RAMdisk and run Windows 7 from the RAMdisk, totally... make an SSD look like it's standing still. :)
 
OS "Awareness" of hyperthreading is crucial for good performance.

I have a 3.0GHz 8 core (16 hyperthreads) Nehalem Mac Pro... and OSX 10.5 is NOT hyper-threading aware and screws it up all the time.

I write massively parallel scientific simulation code for a living... so I get the most out of every core available to me.

The problem with an OS not being aware of hyper-threading is that it will happily schedule two compute intensive tasks to the same physical CPU (one on each of the two hyper-threads the CPU shows up as). This is NON-OPTIMAL when there are other free physical cores laying around not getting used because these two processes will compete for compute resources (registers, floating-point pipelines, L2 cache, etc.) when they could be running happily on separate cores.

Now Nehalem is not as bad in this situation as previous hyper-threaded CPUs were. It has a big, fast shared L3 cache that limits cache misses... and the upgraded memory architecture (finally catching up with AMD) means that the penalty is smaller for cache misses that do happen. BUT.... like I say... you still get performance degradation over running the processes on separate physical cores.

I don't know about Windows (can't use it... since it's not UNIX based)... but Linux does a wonderful job of scheduling processes on hyper-threaded processes. It diligently keeps them off the same physical CPUs when possible.... you can watch it move processes around as they get spawned and settle in on a single CPU.

Like I mentioned earlier OSX 10.5 is TERRIBLE. If I'm running an 8 processor job on my Mac Pro it will happily schedule all 8 processes to _4_ physical CPUs! Even worse... since this is a dual-quadcore machine.... all _8_ of those processes share L3 cache! All while a _whole_ physical CPU sits empty.

I've raised this issue to our Apple representatives and are working with them to get this corrected before Snow Leopard comes out.... let's hope it does....

One more thing.... putting two _threads_ of the same process that are working on the same data on one core can actually _improve_ performance a lot. This is because the data will be sitting in cache for both threads... and it keeps communication down a lot. This works really well in a producer / consumer threading model where one thread is setting up work packages while another is doing computations on those packages. If both of those threads are on the same hyper-threaded core you can see some ridiculous speed increases. In this regard OSX does really well. Their threading model allows you to give hints to the scheduler that two threads will be or won't be using the same data... and the scheduler will try to put them closer / further away from each-other physically.

Anyway... this is a long-winded way of saying that the OS being "aware" of hyper-threading is critical to getting good performance. Microsoft saying that they are building in support for hyper-threading is definitely a good thing...

Friedmud

Lol, I thought that people said OSX WILL JUST WORK ;)
 
Assuming work doesn't provide me with a new machine prior to the fall, it'd be nice to get more life out of my dual-Irwindale Xeon 3GHz. Both chips have have HyperThreading.
Maybe it will help make up for lack of hardware VT, too.
 
I believe that virtual processor such as Hyper-Thread is more hype than anything else.
HT was a originally a patch design by Intel to compete with AMD's Athlon.

Unfortunately, most applications still do not benefit from true multiple cores processors.
Furthermore, in order to benefit from HT, an application must be specifically adapted which increases the development burden. HT may be also detrimental to the performance of some applications. The good news is that HT can be disabled in the Bios.

I would like some numbers to back up the Windows 7 performance boost...
 
I believe that virtual processor such as Hyper-Thread is more hype than anything else.
HT was a originally a patch design by Intel to compete with AMD's Athlon.
While simultaneous multithreading doesn't provide a huge boost for most applications, combined with other techniques it does help to get that very last bit of instruction level parallelism from the processor.

I think the only reason it hasn't made its way to all modern desktop processors is that duplicating register files (reorder buffers, other resources, etc.) for the different process contexts takes up a bit of area. Furthermore, as friedmud pointed out, there's some trickiness in scheduling processes optimally (on behalf of the OS and now the processor).

And I say this as someone who worked for AMD.
 
While simultaneous multithreading doesn't provide a huge boost for most applications, combined with other techniques it does help to get that very last bit of instruction level parallelism from the processor.

I think the only reason it hasn't made its way to all modern desktop processors is that duplicating register files (reorder buffers, other resources, etc.) for the different process contexts takes up a bit of area. Furthermore, as friedmud pointed out, there's some trickiness in scheduling processes optimally (on behalf of the OS and now the processor).

And I say this as someone who worked for AMD.

I assume that's why everything in the new architecture shares that huge L3 cache. Small L1/L2 cache can really save some die real estate.
 
For a 4 or 8 core Xeon workstation with ECC RAM, it is actually among the cheapest out there. Ever since Apple went from G5s to Xeons for the Mac Pros, they have been cheaper than same-specced machines by Dell or HP by $500 to $1000. The problem is that there is a very high base minimum spec with the Mac Pro, while with something like a Dell Precision you can tweak the specs much much lower. The other problem is that most people don't need such a machine. I certainly don't, even though I run Final Cut Studio on my Macs. A Core 2 Duo or consumer level i7 is plenty for me, and IMHO a desktop Mac with that level of CPU with PCI-E expansion is the only real hole in Apple's hardware lineup.

If you need a powerhouse workstation though, its well worth it, otherwise I suspect Apple, Dell, HP, etc, wouldn't bother selling multi-CPU Xeons in the first place.

The Mac Pro is the most compromised workstation on the market IMO. Its more like a desktop with hints of being a workstation. It does look sexy, however.
 
The Mac Pro is the most compromised workstation on the market IMO. Its more like a desktop with hints of being a workstation. It does look sexy, however.

What is you reasoning for this?

The Mac Pro has more "workstation" characteristics than most workstations do. It offers a disk array, ECC memory, Xeons, and Fiber Networking. The only thing differentiating a Dell "workstation" from their "desktops" is a slightly better baseline video card.
 
What is you reasoning for this?

The Mac Pro has more "workstation" characteristics than most workstations do. It offers a disk array, ECC memory, Xeons, and Fiber Networking. The only thing differentiating a Dell "workstation" from their "desktops" is a slightly better baseline video card.

Cooling
Lack of hot swapability (fans, HDs)
No workstation graphics cards available directly from Apple
No SAS drives available from Apple
Proprietary video out connections (albeit one DVI per card is still available) - Many professionals I know use more than 1 monitor, and they are all DVI (granted an extra adapter is available).
8GB memory limit on the standard Mac Pro model (RAM from Apple) - Come on, thats pathetic.
 
Cooling, hot swapping, and bleeding-edge graphics drivers are all desktop enthusiast features, not workstation features, so you really can't complain about that. Once a business spends $20k setting up a workstation, they leave it alone forever.

SAS drives I'll give you, but it's hard to imagine a professional needing that, since SAS is a server drive spec, not for desktop/workstations.

Proprietary video connections, I'm tired of hearing about. DisplayPort is a free/open spec; there's adapters available from Apple right now, and more from 3rd parties on the way.

I'm not sure what you mean on the memory. The 8-core version (the one we are talking about in this thread) goes up to 32GB.
 
Cooling, hot swapping, and bleeding-edge graphics drivers are all desktop enthusiast features, not workstation features, so you really can't complain about that. Once a business spends $20k setting up a workstation, they leave it alone forever.

SAS drives I'll give you, but it's hard to imagine a professional needing that, since SAS is a server drive spec, not for desktop/workstations.

Proprietary video connections, I'm tired of hearing about. DisplayPort is a free/open spec; there's adapters available from Apple right now, and more from 3rd parties on the way.

I'm not sure what you mean on the memory. The 8-core version (the one we are talking about in this thread) goes up to 32GB.

I'm not talking about GTX 295, 4870X2 cards etc. I mean Quadros and FireGLs. The new Mac Pros don't support those from Apple.

For the memory, I mean the stock 4-core 3500 series Xeon can only be outfitted with a maximum of 8GB of ram directly from Apple. The 8-core variant, of course, supports 32GB of ram. From what I've seen on Apple forums, about 50-60% of people buying new Mac Pros are going for the 4-core 2.66 or 2.93GHz models.
 
@ friedmud

The OS has been hyperthreading aware since WinXP. Starting with the Pentium-D, hyperthreading was shown to offer at most about a 10% performance increase. Two physical cores have shown the ability to offer close to 100%. The range that an increase could be, is practically nothing. As always, it's still going to depend on the app to determine whether it makes a difference. This is more BS from MS.
 
@ friedmud

The OS has been hyperthreading aware since WinXP. Starting with the Pentium-D, hyperthreading was shown to offer at most about a 10% performance increase. Two physical cores have shown the ability to offer close to 100%. The range that an increase could be, is practically nothing. As always, it's still going to depend on the app to determine whether it makes a difference. This is more BS from MS.
I'm not entirely sure if WinXP was simultaneous/symmetric multithreading aware so much as it supported multiple cores. Simultaneous multithreading presents a single physical processor to the OS as two, AFAIK no support beyond this is necessary to take advantage of a SMT system in some minimal sense.

As friedmud so interestingly pointed out, there are a number of scheduling issues that come about from SMT: namely, you probably want to distribute separate processes across separate physical cores so as to minimize resource hazards and contention. Whereas with threads (or anything which heavily shares an address space or working data set) it can actually make sense to schedule them on the same physical core, in effect sharing the L1 cache between them.

Basic support for SMT (ie. support for multiple cores) doesn't necessarily mean optimized for SMT. Amusingly enough I'm not even sure if this was what the article was talking about.
 
I've raised this issue to our Apple representatives and are working with them to get this corrected before Snow Leopard comes out.... let's hope it does....

I suspect that you will be waiting for 10.6 for this. This sort of thing is a major bulletpoint feature of the Snow Leopard, doesn't make sense to suddenly add it to 10.5 now that it is so close to end of life.
 
Back
Top