Archives Discussions

[Tristan23]

Now we have 5 (five!) Catalyst versions (11.5 -11.9) with this bad 100% CPU load bug. It was not there in 11.4.

Does AMD not care about it's customers?

This bug affects ALL of us!

I am seriously pi**ed!!

[genaganna]

Originally posted by: Tristan23 Now we have 5 (five!) Catalyst versions (11.5 -11.9) with this bad 100% CPU load bug. It was not there in 11.4.

Does AMD not care about it's customers?

This bug affects ALL of us!

I am seriously pi**ed!! 

In 11.9,  It is fixed for windows and Linux fix will be available in future releases.

[Tristan23]

Originally posted by: genaganna

In 11.9,  It is fixed for windows and Linux fix will be available in future releases.

No, it's not fixed.

As I already wrote for the 11.9 RC driver, for system with more than one GPU the bug still exists!

I have a system with 2 x 6970 - and I get 100% CPU load per CPU. With Catalyst 11.9

Question: Do you not test your software at AMD before you release it????

[Raistmer]

genaganna, did you read my report?
It "fixed" at the cost of unacceptable poor performance. In fact, Cat 11.9 just as unusable in high-performance computations as 11.8. For my own GPU (HD6950) I see degradation on all Catalyst drivers after 11.2 ! It's much more than just 5 releases....

[maximmoroz]

I don't experience any performance degradation with 11.9 (in comparison with 11.8). CPU Load  is about 15% for 4-core CPU.

[Raistmer]

Originally posted by: maximmoroz

I don't experience any performance degradation with 11.9 (in comparison with 11.8). CPU Load  is about 15% for 4-core CPU.

Do you use CPU for computations while GPU working or only to feed GPU?

[maximmoroz]

No, CPU is used only for feeding data to the GPU (data copying is overlapped with kernels running).

[Raistmer]

Originally posted by: maximmoroz

No, CPU is used only for feeding data to the GPU (data copying is overlapped with kernels running).

There is no elapsed time increase under such conditions in my case too.
Only when CPU is busy (even with idle-priority tasks!).

[genaganna]

Originally posted by: Raistmer genaganna, did you read my report? It "fixed" at the cost of unacceptable poor performance. In fact, Cat 11.9 just as unusable in high-performance computations as 11.8. For my own GPU (HD6950) I see degradation on all Catalyst drivers after 11.2 ! It's much more than just 5 releases....

Could you please give a test case to reproduce this performance difference?

Are you using SDK2.5 for both drivers(11.9 and 11.2)?

[Raistmer]

Originally posted by: genaganna

Originally posted by: Raistmer genaganna, did you read my report? It "fixed" at the cost of unacceptable poor performance. In fact, Cat 11.9 just as unusable in high-performance computations as 11.8. For my own GPU (HD6950) I see degradation on all Catalyst drivers after 11.2 ! It's much more than just 5 releases....

Could you please give a test case to reproduce this performance difference?




Are you using SDK2.5 for both drivers(11.9 and 11.2)?

I can provide benchmark, but to see that difference you need to busy CPU with some tasks too. Those tasks can be idle priority ones (so, driver/runtime "should" get all CPU it needs...).

App was compiled on another host (actually it's netbook w/o compatible GPU at all. APP SDK 2.5 was used at compile time). CL file provided in text form, no precompiled kernels.
Thn app's exe + cl file used on another host under Vista x86 + Cat11.2 and Win7x64 Cat 11.9 Guru3D.
So, all app kernels were recompiled under corresponding Catalyst drivers.
And pp executable was linked against APP SDK 2.5 (in both cases).

[genaganna]

Originally posted by: Raistmer

Originally posted by: genaganna

Originally posted by: Raistmer genaganna, did you read my report? It "fixed" at the cost of unacceptable poor performance. In fact, Cat 11.9 just as unusable in high-performance computations as 11.8. For my own GPU (HD6950) I see degradation on all Catalyst drivers after 11.2 ! It's much more than just 5 releases....

Could you please give a test case to reproduce this performance difference?

Are you using SDK2.5 for both drivers(11.9 and 11.2)?

I can provide benchmark, but to see that difference you need to busy CPU with some tasks too. Those tasks can be idle priority ones (so, driver/runtime "should" get all CPU it needs...). App was compiled on another host (actually it's netbook w/o compatible GPU at all. APP SDK 2.5 was used at compile time). CL file provided in text form, no precompiled kernels. Thn app's exe + cl file used on another host under Vista x86 + Cat11.2 and Win7x64 Cat 11.9 Guru3D. So, all app kernels were recompiled under corresponding Catalyst drivers. And pp executable was linked against APP SDK 2.5 (in both cases).

Please give us benchmark.

[Raistmer]

Originally posted by: genaganna


Please give us benchmark.

Here the link to RARed benchmark: MB_bench_v208

The easiest way to reproduce environment for my test is to join to one of BOINC (boinc.berkeley.edu) projects. App in benchmark designed for SETI@home BOINC project. So, it's required to work fast when CPU under 100% load too.
This bench will disable/enable BOINC subsystem automatically. As alternative, one can use some another idle-priority CPU-consuming processes to mimic required environment.

To start testing just run one of CMD files in top subdirectory. also, there is short readme.

[Raistmer]

genaganna, did you reproduce issue?

[Tristan23]

When members of the AMD team write "Please give us more info" it usually means: "We don't care about your s**t - but at least we give you the impression that we would."

[hashman]

It certainly doesn't help when people claim "it is fixed".

Really because every single MULTIPLE-GPU system I have tested the 100% cpu bug remains.

Did a clean install of Windows 7 on machine w/ 2x 5970s, installed Catalyst, installed SDK, ran examples in SDK  .... drumroll ... 100% CPU bug is present.

So to those who are claiming "it is fixed" could you be a little more specific.  Please provide the OS, number of GPU, GPU models, and specific kernel example where the "100% CPU BUG" is "fixed". 

[genaganna]

Originally posted by: hashman It certainly doesn't help when people claim "it is fixed".

Really because every single MULTIPLE-GPU system I have tested the 100% cpu bug remains.

Did a clean install of Windows 7 on machine w/ 2x 5970s, installed Catalyst, installed SDK, ran examples in SDK  .... drumroll ... 100% CPU bug is present. So to those who are claiming "it is fixed" could you be a little more specific.  Please provide the OS, number of GPU, GPU models, and specific kernel example where the "100% CPU BUG" is "fixed". 

Hey guys,

It is fixed in windows but Internally we found few more issues also in windows.  i will let you know more information.

[Tristan23]

genaganna wrote:

It is fixed in windows but Internally we found few more issues also in windows.  i will let you know more information.

OK, so

1) When can we expect a hotfix?

2) When can we expect a final version?

And please don't start bulls**tting us!

[genaganna]

Originally posted by: Tristan23 genaganna wrote:

It is fixed in windows but Internally we found few more issues also in windows.  i will let you know more information.

OK, so

1) When can we expect a hotfix?

2) When can we expect a final version?

And please don't start bulls**tting us!

Tristan23,

I cannot tell you time lines.

You should also remember that we are here to help each other. Not for some thing else.

[FrodoTheGiant]

Originally posted by: genaganna

You should also remember that we are here to help each other. Not for some thing else.

First and foremost: Don't forget who pays your paycheck. That's us, the AMD users (that you obviously give a damn about).

We paid you (AMD) a lot of money for hardware that we can't use properly because of your crappy drivers!

If I don`t have a fix by the end of the month I`ll

1) send my AMD hardware back

2) claim damages

3) buy NVIDIA hardware

I hope other users will do the same.

Possibly this is the only way to make you arrogant %#&!"$ react ...

[Raistmer]

FYI - Cat 11.10 just plain broken. My app produces wrong results with it, all prev releases at least ensure correct results...

[genaganna]

Originally posted by: Raistmer FYI - Cat 11.10 just plain broken. My app produces wrong results with it, all prev releases at least ensure correct results...

Raistmer,

We downloaded the 11.10 preview driver. I did not see issues with the applications i have.

How can i conclude your MB_bench_v208 is runing fine(giving correct results)?     

I see following in log. What is the meaning of theses.

ERROR: The system was unable to find the specified registry key or value

Result : Different

Is it possible to get your application for testing?

[Raistmer]

Originally posted by: genaganna

Originally posted by: Raistmer FYI - Cat 11.10 just plain broken. My app produces wrong results with it, all prev releases at least ensure correct results...

Raistmer,




We downloaded the 11.10 preview driver. I did not see issues with the applications i have.




How can i conclude your MB_bench_v208 is runing fine(giving correct results)?     




I see following in log. What is the meaning of theses.




ERROR: The system was unable to find the specified registry key or value




Result : Different




 




Is it possible to get your application for testing?




 




 

I will provide sample for "all OK" log and for log generated under 11.10preview:

All OK log:


MB_bench_208_CPU_ON.cmd
======================================
4 testWU(s) found
(PG0009_v7.wu)
(PG0395_v7.wu)
(PG0444_v7.wu)
(PG1327_v7.wu)
1 reference science app(s) found
(MB7_win_x86_SSSE3x_CPU_r374.exe -verb -nog)
3 science app(s) found
(MB7_win_x86_SSE3_OpenCL_ATi_r374.exe -period_iterations_num 1 -hp)
(MB7_win_x86_SSE3_OpenCL_ATi_r376.exe -period_iterations_num 1 -hp)
(MB7_win_x86_SSE3_OpenCL_ATi_r376_zero_copy.exe -period_iterations_num 1 -hp)
======================================
MB7_win_x86_SSSE3x_CPU_r374.exe -verb -nog / PG0009_v7.wu :
Result cached, skipping execution
547.371 secs Elapsed
549.077 secs CPU time

Stderr.txt : not found
------------
MB7_win_x86_SSE3_OpenCL_ATi_r374.exe -period_iterations_num 1 -hp / PG0009_v7.wu :
AppName: MB7_win_x86_SSE3_OpenCL_ATi_r374.exe
AppArgs: -period_iterations_num 1 -hp
TaskName: PG0009_v7.wu
Started at : 20:16:03.421
Ended at : 20:18:06.682
123.120 secs Elapsed
34.679 secs CPU time
Speedup : 93.68%
Ratio : 15.83x

R2: .\ref\ref-MB7_win_x86_SSSE3x_CPU_r374.exe-PG0009_v7.wu.res
Result : Strongly similar, Q= 95.09%

R2: .\ref\ref-MB_6.10_win_x86_SSE3_OpenCL_ATi_r246_SETI7.exe-PG0009_v7.wu.res
Result : Strongly similar, Q= 99.91%

R2: .\ref\ref-MB_6.10_win_x86_SSSE3x_CPU_r246_SETI7.exe-PG0009_v7.wu.res
Result : Strongly similar, Q= 95.09%

R2: .\ref\ref-setiathome_6.90_J02UPX_windows_intelx86.exe-PG0009_v7.wu.res
Result : Strongly similar, Q= 95.10%

R2: .\ref\ref-setiathome_6.91_windows_intelx86.exe-PG0009_v7.wu.res
Result : Strongly similar, Q= 95.10%

[ stderr ]

Here results from app under testing compared with already know good ones from prev apps/reference apps. Results similarity expressed in %. No need to go further, just assume >95% is OK for now (different issue - % is lower through last Cat releases it seems, but still acceptable if >95%, lets discuss small issues after fixing of biggest one).

Here the log under Cat11.10 preview:


MB_bench_208_CPU_ON.cmd
======================================
4 testWU(s) found
(PG0009_v7.wu)
(PG0395_v7.wu)
(PG0444_v7.wu)
(PG1327_v7.wu)
1 reference science app(s) found
(MB7_win_x86_SSSE3x_CPU_r374.exe -verb -nog)
3 science app(s) found
(MB7_win_x86_SSE3_OpenCL_ATi_r374.exe -period_iterations_num 1 -hp)
(MB7_win_x86_SSE3_OpenCL_ATi_r376.exe -period_iterations_num 1 -hp)
(MB7_win_x86_SSE3_OpenCL_ATi_r376_zero_copy.exe -period_iterations_num 1 -hp)
======================================
MB7_win_x86_SSSE3x_CPU_r374.exe -verb -nog / PG0009_v7.wu :
Result cached, skipping execution
547.371 secs Elapsed
549.077 secs CPU time

Stderr.txt : not found
------------
MB7_win_x86_SSE3_OpenCL_ATi_r374.exe -period_iterations_num 1 -hp / PG0009_v7.wu :
AppName: MB7_win_x86_SSE3_OpenCL_ATi_r374.exe
AppArgs: -period_iterations_num 1 -hp
TaskName: PG0009_v7.wu
Started at : 23:04:22.491
Ended at : 23:05:00.726
38.048 secs Elapsed
33.462 secs CPU time
Speedup : 93.91%
Ratio : 16.41x

R2: .\ref\ref-MB7_win_x86_SSSE3x_CPU_r374.exe-PG0009_v7.wu.res
----- R1:R2 ------ ----- R2:R1 ------
Tight Good Bad Tight Good Bad
Spike 0 0 0 0 0 0
Autocorr 0 0 30 0 0 0
Gaussian 0 0 0 0 0 0
Pulse 0 0 0 0 0 1
Triplet 0 0 0 0 0 0
Best Spike 0 0 0 0 0 1
Best Autocorr 0 0 0 0 0 1
Best Gaussian 0 0 0 0 0 1
Best Pulse 0 0 0 0 0 1
Best Triplet 0 0 0 0 0 0
---- ---- ---- ---- ---- ----
0 0 30 0 0 5

Unmatched signal(s) in R1 at line(s) 671 688 705 722 739 756 773 790 807 824 841 858 875 892 909 926 943 960 977 994 1011 1028 1045 1062 1079 1096 1113 1130 1147 1164
Unmatched signal(s) in R2 at line(s) 671 698 714 731 752
Result : Different.

R2: .\ref\ref-MB_6.10_win_x86_SSE3_OpenCL_ATi_r246_SETI7.exe-PG0009_v7.wu.res
----- R1:R2 ------ ----- R2:R1 ------
Tight Good Bad Tight Good Bad
Spike 0 0 0 0 0 0
Autocorr 0 0 30 0 0 0
Gaussian 0 0 0 0 0 0
Pulse 0 0 0 0 0 1
Triplet 0 0 0 0 0 0
Best Spike 0 0 0 0 0 1
Best Autocorr 0 0 0 0 0 1
Best Gaussian 0 0 0 0 0 1
Best Pulse 0 0 0 0 0 1
Best Triplet 0 0 0 0 0 0
---- ---- ---- ---- ---- ----
0 0 30 0 0 5

Unmatched signal(s) in R1 at line(s) 671 688 705 722 739 756 773 790 807 824 841 858 875 892 909 926 943 960 977 994 1011 1028 1045 1062 1079 1096 1113 1130 1147 1164
Unmatched signal(s) in R2 at line(s) 671 698 714 731 752
Result : Different.

R2: .\ref\ref-MB_6.10_win_x86_SSSE3x_CPU_r246_SETI7.exe-PG0009_v7.wu.res
----- R1:R2 ------ ----- R2:R1 ------
Tight Good Bad Tight Good Bad
Spike 0 0 0 0 0 0
Autocorr 0 0 30 0 0 0
Gaussian 0 0 0 0 0 0
Pulse 0 0 0 0 0 1
Triplet 0 0 0 0 0 0
Best Spike 0 0 0 0 0 1
Best Autocorr 0 0 0 0 0 1
Best Gaussian 0 0 0 0 0 1
Best Pulse 0 0 0 0 0 1
Best Triplet 0 0 0 0 0 0
---- ---- ---- ---- ---- ----
0 0 30 0 0 5

Unmatched signal(s) in R1 at line(s) 671 688 705 722 739 756 773 790 807 824 841 858 875 892 909 926 943 960 977 994 1011 1028 1045 1062 1079 1096 1113 1130 1147 1164
Unmatched signal(s) in R2 at line(s) 671 698 714 731 752
Result : Different.

R2: .\ref\ref-setiathome_6.90_J02UPX_windows_intelx86.exe-PG0009_v7.wu.res
----- R1:R2 ------ ----- R2:R1 ------
Tight Good Bad Tight Good Bad
Spike 0 0 0 0 0 0
Autocorr 0 0 30 0 0 0
Gaussian 0 0 0 0 0 0
Pulse 0 0 0 0 0 1
Triplet 0 0 0 0 0 0
Best Spike 0 0 0 0 0 1
Best Autocorr 0 0 0 0 0 1
Best Gaussian 0 0 0 0 0 1
Best Pulse 0 0 0 0 0 1
Best Triplet 0 0 0 0 0 0
---- ---- ---- ---- ---- ----
0 0 30 0 0 5

Unmatched signal(s) in R1 at line(s) 671 688 705 722 739 756 773 790 807 824 841 858 875 892 909 926 943 960 977 994 1011 1028 1045 1062 1079 1096 1113 1130 1147 1164
Unmatched signal(s) in R2 at line(s) 671 698 714 731 752
Result : Different.

R2: .\ref\ref-setiathome_6.91_windows_intelx86.exe-PG0009_v7.wu.res
----- R1:R2 ------ ----- R2:R1 ------
Tight Good Bad Tight Good Bad
Spike 0 0 0 0 0 0
Autocorr 0 0 30 0 0 0
Gaussian 0 0 0 0 0 0
Pulse 0 0 0 0 0 1
Triplet 0 0 0 0 0 0
Best Spike 0 0 0 0 0 1
Best Autocorr 0 0 0 0 0 1
Best Gaussian 0 0 0 0 0 1
Best Pulse 0 0 0 0 0 1
Best Triplet 0 0 0 0 0 0
---- ---- ---- ---- ---- ----
0 0 30 0 0 5

Unmatched signal(s) in R1 at line(s) 671 688 705 722 739 756 773 790 807 824 841 858 875 892 909 926 943 960 977 994 1011 1028 1045 1062 1079 1096 1113 1130 1147 1164
Unmatched signal(s) in R2 at line(s) 671 698 714 731 752
Result : Different.

[ stderr ]

Many unmatched results.
For you all what is matters - results were invalid. If you see this - something is wrong.

About registry setting error issue - safely ignore it. It should disable BOINC but you obviously have no BOINC installed, just continue to app run itself.
Current app versions located inder Science_apps directory. You not need running all of them (I provided few to discuss another issue later, looks like ZERO-copy path not giving any benefits but again, lets solve show-stopper problem first).

[Raistmer]

Shorter version:
If you see "Result : Strongly similar, Q= xxx% " - app worked correct.
If you see: "Result : Different. " - something was wrong.

Also, keep in mind that log also stored under Testdatas directory. One can always check run later.