AMD 'Bulldozer' gets an Update from Microsoft

[Qianglong]

The download link seems invalid on the MS hotfix page...where else can I get this patch guys?

 

[Nemesis 1]

http://www.tomshardware.com/reviews/fx-8150-zambezi-bulldozer-990fx,3043-23.html

Toms did a preview on Win8 developer, only 2 applications tested but performance did improve over Win7. It seems they actually put more threads on the same module when possible so other modules and deactivate allowing higher turbo speeds. If the updates to Win7 scheduler function the same way, I don't expect to see much if any improvement for those who disable turbo core and overclock. It is interested FX gets slightly improved performance in Win8 while Sandy Bridge lost performance.

Yes it is interesting the SB looses performance .Infact its grounds for a law suite class action time again.

 

[Nemesis 1]

Umm, Intel did need "special microsoft updates" when they released their original HT CPUs.

The only issue here is why AMD didn't work with Microsoft to get this released prior BD's launch. The scheduler is one of the fundamental functions of a modern OS, and while the changes needed for BD should have been relatively minor, AMD should have known that MS wasn't going to ship a hotfix to the scheduler without serious validation.

Yes it did and MS dragged their feet on that update.

 

[Idontcare]

so, how in the world didn't AMD get Microsoft to implement this by launch date? Failure...

Its business. My guess is Microsoft said "we can have it ready in 18 months for $, we can have it ready in 12 months for $$, or we can have it ready in 6 months for $$$$...which are you feeling flush enough to pay for?"

 

[IntelUser2000]

Yes it is interesting the SB looses performance .Infact its grounds for a law suite class action time again.

It seems in the other thread 2600K gained performance, so the new patch and Windows 8 gives slight boost to Hyperthreading and module based devices while lowering performance a bit on the core-only chips.

Maybe its a conspiracy to lower performance on non-logical thread enabled chips so the logical enabled thread ones look better.

 

[Cerb]

so, how in the world didn't AMD get Microsoft to implement this by launch date? Failure...

AMD and MS would need to work together on it, and to be sure it's correct, MS would need chip revisions guaranteed to give exactly the same performance as final production models. If they had it ready on the day of launch, they would still need to spend time testing it, and likely would not have gotten launch-equivalent CPU samples more than a few weeks ahead of the rest of the world.

Reliability anomalies are somewhat rare, thankfully, but performance anomalies are common, and have been getting patched for a good long while, either when MS or the vendor gets around to it.

It's more impressive on Intel's part that Nehalem and SB did not need any significant performance patching, than it is a black mark on AMD that BD needs it.

 

[Dresdenboy]

It's more impressive on Intel's part that Nehalem and SB did not need any significant performance patching, than it is a black mark on AMD that BD needs it.

I think, either my principles of logic are leaving me or someone else alone. There is this "no need" posting above and a parallel thread "AMD hot fix just boosted my sandy in cinebench".

The SMT principle on Intel's side is known since P4 Northwood. Treating the CPU accordingly helps, if there are more than one physical core. But treating 2 cores with shared resources as completely independent is a problem.

So which FAIL is FAIL and which FAIL is no FAIL now?

 

[hans030390]

Nice fine. I'm really liking my FX 6100 and Asus M5A97 setup. Got the CPU for $100 on Black Friday and the motherboard for 20% off (came to $80, also Black Friday). I don't think you can get a much better deal than that, and it's a great upgrade from my PII X3 720. Got an aftermarket cooler on there and have it running at 4.3GHz right now (overvolted). I'm planning on upgrading to a better cooler so that I can push it higher. Realistically, with the amount of time this computer spends idle, I'm not worried so much about power consumption.

 

[denev2004]

It's more impressive on Intel's part that Nehalem and SB did not need any significant performance patching, than it is a black mark on AMD that BD needs it.

I guess they do need one...talking about patch for AVX, which I think is a part of Windows 7 SP1.
Well, I know is kind of usual for extend instruction to have a patch

 

[Riek]

It's more impressive on Intel's part that Nehalem and SB did not need any significant performance patching, than it is a black mark on AMD that BD needs it.

Actually as previously mentionned HT did require OS updates, without them SB would be magnitudes slower in 2 - 4thread combinations (heck it wouldn't even be competitive with current BD in that range).
But you are correct that AMD could simulate HT like design easily with a HT flag on the second int core. Which would results in first populating over modules before populating the cores in the modules. (but that is not what they wanted though)

Yes it is interesting the SB looses performance .Infact its grounds for a law suite class action time again.

sure thing... lol. Even on a non - 1 april fools day people won't take that quote seriously.

 

[Deanodarlo]

Remember all those arguements if Bulldozer was a 8 core or 4 core, where an AMD Module = Intel core.

With this patch, it now shows 4 cores 8 threads doesn't it?

So better performance by treating a module as a single core.

 

[Cerb]

I think, either my principles of logic are leaving me or someone else alone. There is this "no need" posting above and a parallel thread "AMD hot fix just boosted my sandy in cinebench".

http://www.anandtech.com/bench/Product/191?vs=49
http://www.anandtech.com/bench/Product/288?vs=191
Clear improvement across the board.

http://www.anandtech.com/bench/Product/434?vs=203
http://www.tomshardware.com/reviews/fx-8150-zambezi-bulldozer-990fx,3043-20.html
Blah.

http://www.tomshardware.com/reviews/fx-8150-zambezi-bulldozer-990fx,3043-6.html
Ew.

SB was, without any fixing, far superior to Stars and Nehalem. Nehalem was far superior to Conroe and friends, and also Stars. As such, it is not in need. BD ranges from significantly worse to better than its predecessors and and competition without fixing. Typically it is the same or a bit worse, if you can't pack on the threads, and shows much more regular performance scaling under Linux than under Windows. So, it can be seen as needing some fixing.

 

[Riek]

http://www.anandtech.com/bench/Product/191?vs=49
http://www.anandtech.com/bench/Product/288?vs=191
Clear improvement across the board.

http://www.anandtech.com/bench/Product/434?vs=203
http://www.tomshardware.com/reviews/fx-8150-zambezi-bulldozer-990fx,3043-20.html
Blah.

http://www.tomshardware.com/reviews/fx-8150-zambezi-bulldozer-990fx,3043-6.html
Ew.

SB was, without any fixing, far superior to Stars and Nehalem. Nehalem was far superior to Conroe and friends, and also Stars. As such, it is not in need. BD ranges from significantly worse to better than its predecessors and and competition without fixing. Typically it is the same or a bit worse, if you can't pack on the threads, and shows much more regular performance scaling under Linux than under Windows. So, it can be seen as needing some fixing.

So a cpu's that brings nothing new technoology wise doesn't need fixing and a cpu that brings something entirely new does need fixing... how is this supposed to be illogical, bad?

If HT wasn't fixed by p4 and during the nehalem time, SB would need a fix to compete with BD... without the fix. (since the performance would drop with 70% for 2 threads....)

 

[pelov]

OMG!
i just can't belive this

http://www.overclock.net/t/1185039/the-1st-windows-7-scheduler-patch-results-are-in-have-fun/60#post_15955453

The most drastic improvement you'll see is tasks ranging from 2-4 threads, and since MW3 sees Bulldozer as a 4 core processor now (4 modules) it performs significantly better. It's not that surprising. The single-threaded benchmarks have stayed pretty much the same from the looks of it, but those weren't going to change.

I'm interested to see just how this scheduler fix approaches and recognizes their design. Some apps are showing it as a 4 core 8 thread and others as 8 core 8 thread, which makes sense given that due to the architecture the number of cores/threads depends on how the CPU is being tasked. On the other hand, it also seems to have improved hyperthreading as well =P

I really hope AMD does deliver more than their estimated performance boost with Piledriver. Even with an improved windows scheduler they've still got a long way to go

 

[dma0991]

Microsoft Pulls Down the AMD Bulldozer Multi-Threaded Patch

The patch got pulled down for unknown problems. D: Someone should re upload it elsewhere just in case someone else needs it.

 

[formulav8]

So, it can be seen as needing some fixing.

Huh? SB is the typical cpu, no big difference. BD is Very different, complicated, not the norm, new. So of course there will be teething issues. Even HT needed Winxp SP1 to work. Get over it. :whiste:

 

[Idontcare]

Microsoft Pulls Down the AMD Bulldozer Multi-Threaded Patch

The patch got pulled down for unknown problems. D: Someone should re upload it elsewhere just in case someone else needs it.

They forgot to set the "check CPUID" switch so that the patch only works on AMD systems and not on Intel one's :twisted:

Don't worry, once they fix the "bug" it will be re-released just as Nvidia did with their drivers when they accidentally enabled Physx to run when AMD graphics cards were present along with an Nvidia card for the physx portion.

 

[notty22]

There are subtle differences in what Microsoft calls updates and hotifxes. I know some decided to install this on their rig, AMD or not. IMO, something that makes changes to the Kernel, I've learned not to be the first in line to 'test' these things.

 

[Cerb]

Huh? SB is the typical cpu, no big difference. BD is Very different, complicated, not the norm, new. So of course there will be teething issues. Even HT needed Winxp SP1 to work. Get over it. :whiste:

I believe I've spent at least three posts ITT explaining that Windows patches for SMP performance have been business as usual since the home versions had to care about it. That I've been over.

I was further explaining why this particular one gets so much attention, and why boosting performance for SB is different than boosting performance for BD: that without said patching, it's not a clear successor to Stars; where Intel's recent CPUs have been clearly better than the last, without additional software help.

So a cpu's that brings nothing new technoology wise doesn't need fixing and a cpu that brings something entirely new does need fixing... how is this supposed to be illogical, bad?

It's not. It's a new CPU that typically will not run a single thread of a desktop program as fast as the last generation, when that is an important thing to do.

http://www.tomshardware.com/reviews/fx-8150-zambezi-bulldozer-990fx,3043-6.html
A scheduler fix isn't going to take care of that. I expect multithreaded games to be doing much better, though (once the patch comes back out).

I'm not saying there shouldn't be such patching, but rather, that AMD should have made a good mass market CPU that would rarely, if ever, let Stars appear as good or better at each thread. If they had, it would be a much more minor issue.

 

[nsdjoe]

Am I correct in reading that this patch (assuming it will be re-released) improves SB performance? Only hyperthreaded chips or will my bf3-bottlenecked 2500k see a benefit?

 

[Vesku]

They forgot to set the "check CPUID" switch so that the patch only works on AMD systems and not on Intel one's :twisted:

Don't worry, once they fix the "bug" it will be re-released just as Nvidia did with their drivers when they accidentally enabled Physx to run when AMD graphics cards were present along with an Nvidia card for the physx portion.

It would be pretty hilarious if AMD could get MS to do this, assuming it didn't escalate from there. It would stand up well to the logic I've heard of "they paid for it so it's cool if they lock it down super tight, if you don't like it then buy something else". Year of the Linux Desktop coming? ROFL

 

[sandorski]

They forgot to set the "check CPUID" switch so that the patch only works on AMD systems and not on Intel one's :twisted:

Don't worry, once they fix the "bug" it will be re-released just as Nvidia did with their drivers when they accidentally enabled Physx to run when AMD graphics cards were present along with an Nvidia card for the physx portion.

Possibly. If that's the case, it likely would be to prevent issues with untested(Intel cpu's in this case) hardware.

 

[EightySix Four]

Remember all those arguements if Bulldozer was a 8 core or 4 core, where an AMD Module = Intel core.

With this patch, it now shows 4 cores 8 threads doesn't it?

So better performance by treating a module as a single core.

My understanding is that this is not true. In fact it is the other way around. By treating them as real cores instead of hyperthreading it boosts performance. We'll say there are 4 cores, and an a and b thread for each core. So core 1a, core 1b, core 2a, etc. On a Intel processor the "b" cores are hyperthreaded so the scheduler will want to avoid those until last resort because 1a, 2a, 3a, and 4a are faster. In Bulldozer, if there are two threads, it is faster to put them on 1a and 1b so the other cores clock down and the turbo kicks in.

 

[jones377]

My understanding is that this is not true. In fact it is the other way around. By treating them as real cores instead of hyperthreading it boosts performance. We'll say there are 4 cores, and an a and b thread for each core. So core 1a, core 1b, core 2a, etc. On a Intel processor the "b" cores are hyperthreaded so the scheduler will want to avoid those until last resort because 1a, 2a, 3a, and 4a are faster. In Bulldozer, if there are two threads, it is faster to put them on 1a and 1b so the other cores clock down and the turbo kicks in.

That's the theory, however from earlier tests with forced thread affinity set in windows it's not always the case. You do however get much better power efficiency if you put all threads on as few modules as possible.

 

[thilanliyan]

Apparently MS pulled the patch because there are 2 parts that are needed and was supposed to be released early next year:
http://www.brightsideofnews.com/new...r-patch-by-mistake2c-incomplete-download.aspx

Oh boy...I hope the patch isn't delayed as well.