FX-8000 CPU owners plz help me

[Sabotaged_Enigma]

Lately I've been confused about the floating-point performance of 4 module 8 thread FX-8000 series CPUs.
Bulldozer architecture's theoretical FP32 and FP64 performance are 16 and 8 GFLOPS/(GHz·FPU), and thus a manually overclocked 4.9 GHz FX-8350 should produce a result of 313.6 GFLOPS and 156.8 GFLOPS. But all across the Internet there's no result that's even close.
This could be caused by Windows 7 not properly recognising Bulldozer's module design, since FX-4200 doesn't have this problem.
Although I have proved that FX-4200 is a 4 module 4 thread CPU (link: https://www.techpowerup.com/forums/threads/anyone-remembered-fx-4200.316590/), I haven't got a FX-8000 series CPU, and thus I can't prove it myself, no matter right or wrong.

If you've got a FX-8000 series CPU, and it's convenient for you run AIDA64 GPGPU benchmark, please post the result here. I appreciate your help!
Requirement: Windows 10 64-bit and AIDA64 6.xx and later versions.

 

[DaemonForce]

I really don't feel like setting this up with my RX 580, a Windows 10 desktop install, main power distribution or water cooling assembly and raising the overclock on a Gigglebyte just for one test. The test versioning doesn't even fit any real standard either.

Windows Server 2016
FX-8370 ↗ 4.1
Gigabyte GA-970A-DS3P
AIDA64 v6.20.x
High Performance mode

Honestly I wouldn't get into this expecting full marks as the whole legal screeching over "cores" has everything to do with the "modules" in which floating point calculations are shared. Also I've never attempted to run Windows 7 on FX nor do I care about it as I have more modern functions in mind even if it's now just a glorified network controller with disk shares. I was on Phenom II X4 from 2009-2018 and switched up entirely to mitigate the serious frame issues I experienced in VR. That lasted about a year before going Ryzen, which for clarity:

Does not approach theoretical expectation of FX either.

 

[Vya Domus]

You'll never get the maximum expected throughput on any architecture.

 

[QuietBob]

This is from a very recent version of AIDA with an FX-8300 @ 4.5 GHz all-core on Win7 x64:

 

[Sabotaged_Enigma]

You'll never get the maximum expected throughput on any architecture.

It's true but FX-4200 does achieve full throughput...
The problem is that ratio being about 13.66 is far more than margin of error and thus I need to find prove...

This is from a very recent version of AIDA with an FX-8300 @ 4.5 GHz all-core on Win7 x64:

View attachment 343027

Thank you! This is the same "underachieved" as the results I found on the web

Windows Server 2016
FX-8370 ↗ 4.1
Gigabyte GA-970A-DS3P
AIDA64 v6.20.x
High Performance mode

View attachment 343020

Thank you!

 

[Vya Domus]

It's true but FX-4200 does achieve full throughput

It doesn't, if that CPU has 4 modules it also has the same FP performance as this one, so neither are getting the full throughput.

 

[Deleted member 239770]

It's impossible to reach true throughput whilst running Windows....

 

[Sabotaged_Enigma]

It's impossible to reach true throughput whilst running Windows....

Actually most processors including FX-4200 do. At least it should be obviously margin of error.

It doesn't, if that CPU has 4 modules it also has the same FP performance as this one, so neither are getting the full throughput.

I think you've read my post about FX-4200 because you replied several times there. So you should've known what I'm talking about. If not, let me make it clear for you. I'm talking about whether it's hardware failure or problems caused by bad OS scheduling. Because 13.66 GFLOPS per GHz x FPU is far below than theoretical 16, which is not a matter of "any architecture being impossible to achieve full throughput".
This could be caused by multiple factors, e.g. architecture design fault, OS version, and software version. All I need is actual proof to tell WHICH factor it is.

Come on Bulldozer owners, more proof is needed...
A result of a fully-functioning FX-8000 running AIDA64 7.00 or later versions on Windows 10 x64 or later versions would perfectly settle the problem once and for all!

 

[Melvis]

Did you apply the patch? that is required to FX CPU's for windows 7 and under, maybe this might be the issue? it fixes alot of Performance issues and schedule issues.

 

[Wirko]

This could be caused by multiple factors, e.g. architecture design fault, OS version, and software version. All I need is actual proof to tell WHICH factor it is.

I would also try to underclock the cores as low as practically possible, in order to make all (or most) parts of the system *relatively* faster.

For me, that's one of the principal guidelines for troubleshooting anything. If you don't know how to improve something, try to make it a bit worse. With some luck, that will give you some additional insight into causes and effects.

 

[Sabotaged_Enigma]

Did you apply the patch? that is required to FX CPU's for windows 7 and under, maybe this might be the issue? it fixes alot of Performance issues and schedule issues.

Yeah I know about the patch but never manually applied it. Does Windows 7 Ultimate 7601 come with the patch? I used the final ISO file from Microsoft to install Windows 7.
The bigger problem is that I haven't got a FX-8000 series CPU, so I can't test it...

 

[NastyDjango]

only trial version -.- FX 8350 - 32 GB Ram - Win11 24H2

 

[Fouquin]

Yeah I know about the patch but never manually applied it. Does Windows 7 Ultimate 7601 come with the patch? I used the final ISO file from Microsoft to install Windows 7.
The bigger problem is that I haven't got a FX-8000 series CPU, so I can't test it...

The patch is for x87 code paths, not SSE or AVX. AIDA64 leverages AVX on Bulldozer for the GPGPU tests. It wouldn't help any.

 

[Sabotaged_Enigma]

only trial version -.- FX 8350 - 32 GB Ram - Win11 24H2



View attachment 343317

Thank you! I think this would conclude the problem! It seems that a 4 module 8 thread just could give full FP32 and FP64 performance, but other Bulldozer like 2M/4T, 4M/4T, and 3M/6T could.

The patch is for x87 code paths, not SSE or AVX. AIDA64 leverages AVX on Bulldozer for the GPGPU tests. It wouldn't help any.

Thank you!

FX 8350 - 32 GB Ram - Win11 24H2

By the way, I would like to know how Windows 11 reports FX-8350 in the task manager, is it 8 cores 8 threads or 4 modules 8 threads?
Windows 7 reports 8 cores 8 threads while Windows 10 reports 4 modules 8 threads.

 

[NastyDjango]

 

[Melvis]

Yeah I know about the patch but never manually applied it. Does Windows 7 Ultimate 7601 come with the patch? I used the final ISO file from Microsoft to install Windows 7.
The bigger problem is that I haven't got a FX-8000 series CPU, so I can't test it...

An update is available for computers that have an AMD FX, AMD Opteron 4200, AMD Opteron 6200, or AMD Bulldozer series processor installed and that are running Windows 7 or Windows Server 2008 R2 - Microsoft Support

Describes an update for computers that are running Windows 7 or Windows Server 2008 R2 and have an AMD FX series, AMD Opteron 4200 series, or AMD Opteron 6200 series processor installed. This update improves the system performance when you run multithreaded applications on the computer.

support.microsoft.com

An update that selectively disables the Core Parking feature in Windows 7 or in Windows Server 2008 R2 is available - Microsoft Support

Describes an update that enables the hotfix installer to selectively disable the Core Parking feature in Windows 7 or in Windows Server 2008 R2. The Core Parking feature is disabled if an AMD FX, AMD Opteron™ 4000 Series, or AMD Opteron™ 6000 Series processor is installed.

support.microsoft.com

 

[RJARRRPCGP]

By the way, I would like to know how Windows 11 reports FX-8350 in the task manager, is it 8 cores 8 threads or 4 modules 8 threads?
Windows 7 reports 8 cores 8 threads while Windows 10 reports 4 modules 8 threads.

IIRC, Windows 8.1 and later, in appearance, will look just like SMT, would say "4 cores 8 threads". Zambezi (early-gen) and Vishera (refresh and late-production-runs) are not SMT!
Ryzen was the first to use SMT.

 

[Sabotaged_Enigma]

View attachment 343379

Thank you!

An update is available for computers that have an AMD FX, AMD Opteron 4200, AMD Opteron 6200, or AMD Bulldozer series processor installed and that are running Windows 7 or Windows Server 2008 R2 - Microsoft Support

Describes an update for computers that are running Windows 7 or Windows Server 2008 R2 and have an AMD FX series, AMD Opteron 4200 series, or AMD Opteron 6200 series processor installed. This update improves the system performance when you run multithreaded applications on the computer.

support.microsoft.com

An update that selectively disables the Core Parking feature in Windows 7 or in Windows Server 2008 R2 is available - Microsoft Support

Describes an update that enables the hotfix installer to selectively disable the Core Parking feature in Windows 7 or in Windows Server 2008 R2. The Core Parking feature is disabled if an AMD FX, AMD Opteron™ 4000 Series, or AMD Opteron™ 6000 Series processor is installed.

support.microsoft.com

Thank you!

 

[Wirko]

IIRC, Windows 8.1 and later, in appearance, will look just like SMT, would say "4 cores 8 threads". Zambezi (early-gen) and Vishera (refresh and late-production-runs) are not SMT!
Ryzen was the first to use SMT.

The term for AMD's implementation was CMT, or Clustered Multi-Threading, right? And it used to be described as something that's actually an opposite of SMT.

 

[Gica]

Lately I've been confused about the floating-point performance of 4 module 8 thread FX-8000 series CPUs.

The FX series has a floating point module for two "cores". Performance is disastrous when both cores call this module.
I was very surprised when an FX 8300 could not run Virtual Dub (it uses only two threads) and a game in parallel. Yes, it was running, but the game recorded the biggest CPU freeze I've ever seen.
I was even more surprised when a Pentium G4560 (2 cores, 4 threads, 3.5GHz) destroys it in games.

Probably the biggest fail in the history of processors.

 

[Sabotaged_Enigma]

Performance is disastrous when both cores call this module.
I was very surprised when an FX 8300 could not run Virtual Dub (it uses only two threads) and a game in parallel. Yes, it was running, but the game recorded the biggest CPU freeze I've ever seen.

That's why I'm confused about it.
A 4 module 4 thread FX-4200 doesn't have any strange issue. A 3 module 6 thread FX-6000 model gives full FP performance in GPGPU. ONLY the full-spec 4M8T FX-8000 behave strangely. I'm trying to find out, but very little evidence has been found so far.

 

[Dr. Dro]

That's why I'm confused about it.
A 4 module 4 thread FX-4200 doesn't have any strange issue. A 3 module 6 thread FX-6000 model gives full FP performance in GPGPU. ONLY the full-spec 4M8T FX-8000 behave strangely. I'm trying to find out, but very little evidence has been found so far.

However, an FX-4000 series processor has two modules and four cores.

"Threading" as in SMT/HyperThreading is not a concept with AMD FX, it's a CMT design. In the Heavy Equipment line (Bulldozer through Excavator), one cluster is comprised of two independent integer units that share one floating-point unit. Windows assumes that every core has a floating-point unit of its own (this is valid for every processor architecture EXCEPT for AMD Bulldozer and its derivatives), this is why things such as the Task Manager say that it's a 4-core, 8-thread processor. This information is incorrect.

The reason you'll never see full throughput out of these chips is because this is intentional, it's a design compromise that AMD made at the time. Factor in scalability with core count due to Amdahl's law and interconnect speeds, as well as the dreadful memory controller that these processors have, it's no wonder that they underperformed as badly as they did. It was a gamble that AMD very much lost.

Anyway, I'm of the firm belief that FX are true 8 core processors and AMD just chose to settle as litigious resolution of the dispute would cost far too much, the courts do not understand computers after all.

 

[Sabotaged_Enigma]

However, an FX-4000 series processor has two modules and four cores.

Ha, FX-4200 is the most special among them. I have proved that it is 4M/4T. You can find the link at the top of this thread

 

[Dr. Dro]

Ha, FX-4200 is the most special among them. I have proved that it is 4M/4T. You can find the link at the top of this thread

Are you sure the software you are using to read this is correct? Or it could be an oddball one-off like the Ryzen 5 1600's that came with 8 cores.

https://www.cpu-world.com/CPUs/Bulldozer/AMD-FX-Series%20FX-4200%20(FD4200FRW4KGU).html

CPU-World data clearly shows 2 modules

 

[Sabotaged_Enigma]

Are you sure the software you are using to read this is correct? Or it could be an oddball one-off like the Ryzen 5 1600's that came with 8 cores.

https://www.cpu-world.com/CPUs/Bulldozer/AMD-FX-Series%20FX-4200%20(FD4200FRW4KGU).html

CPU-World data clearly shows 2 modules

Yeah, performance tests all indicate that FX-4200 is different from other FX-4000 series CPUs lol. Rumour had it right around its launch time, but Bulldozer sucked so nobody would bother.
No database records it properly haha