AMD 12.1 previe driver performance

[Raistmer]

Well, in existing apps I relied on OS/driver to cleanup GPU state when process terminated.
It's possible that new AMD drivers can't cleanup properly and need help from application.
So I will not go further in optimization and implement full OpenCL objects cleanup inside app instead. Maybe this could help with this issue.

Probably that'll be along the right lines.  Similar (with slighlty different symptoms) was originally happening under Cuda, resolved by making sure safe exit shutdown code was used, rather than terminating the process/thread during mid kernel execution or memory transfers.  The side effects on Cuda devices/drivers were more along the lines of sending the driver into failsafe/sticky-downclock, but I can see that similar mechanisms might result in resource leaks or other unpredictable problems like that.

Jason

Need to note, there is no talk about unexpected app termination where async PCIe transfers or kernel execution can take place. I speak about regular app ending where all work is done (no async oparations pending). I just don't deallocate buffers and free queue/context objects cause at process termination it supposed to be OS task - to free process resourses on exit. Of course OS can't free GPU resourses w/o help of corresponding driver. And here can be weak place I hope to strengthen soon.

[Raistmer]

One ATI_r521 AP task running slowly, then GPUs suspended and resummed



Claggy

It's just my case. Very low GPU usage that can be improved after restart.

[Raistmer]

well, many hours running in 3 tasks config already - no GPUload drop.
Always ~99%.
For some reasons x1 and x2 configs unstable and x3 stable? Very strange...
Well, when I plot graphs it will be more clear if elapsed time deviation reduced too in x3 config or just GPU load improved for some of 3 tasks and not for some others...

[Mike]

well, many hours running in 3 tasks config already - no GPUload drop.
Always ~99%.
For some reasons x1 and x2 configs unstable and x3 stable? Very strange...
Well, when I plot graphs it will be more clear if elapsed time deviation reduced too in x3 config or just GPU load improved for some of 3 tasks and not for some others...

Thats what i fear Raistmer.

[Claggy]

On GTX460 during NV_r365 benchrun, if i resume Boinc CPU use (GPU use suspended), GPU use drops slightly, then drops to ~Zero after another 5 secs or so:



Without doing anything else, GPU use goes back to 98%, before dropping back down again:


Claggy

[Raistmer]

So, this issue is common for both drivers.
What about -hp switch in menchmark command line? Does it helps ?

[Claggy]

So, this issue is common for both drivers.
What about -hp switch in menchmark command line? Does it helps ?

Setting the -hp switch with NV_r365 didn't help, but leaving a core free does help (in conjuction with -hp switch), but GPU load is only ~88% now instead of ~98%

Starting the Bench without the -hp switch and a core free also gives ~88% GPU load.

Has there been any Microsoft updates that might have interfered with scheduling/thread priority?

Claggy

[Raistmer]

I don't know about such windows updates.
PErformance drop was reported more than month ago, bug was confirmed... and no more info from NV side.

[Claggy]

I don't know about such windows updates.
PErformance drop was reported more than month ago, bug was confirmed... and no more info from NV side.

I saw some of your posts in the Regular Nvidia Forums, have you tried Registering/Posting in their Developer Forums?:

http://forums.developer.nvidia.com/

Claggy

[Raistmer]

No I used bug reporting system only as registered developer. Will try those forums too.