Message boards :
Number crunching :
NVidia OpenCL BUG still exist?
Message board moderation
Author | Message |
---|---|
Sutaru Tsureku Send message Joined: 6 Apr 07 Posts: 7105 Credit: 147,663,825 RAC: 5 |
From the messages here in the forum .. 1 NV OpenCL AP GPU WU use 1 CPU-thread (100 % usage of CPU-thread). With the newest NVidia driver also, the BUG still exist? With all GPU series and operation systems? Thanks. * Best regards! :-) * Philip J. Fry, team seti.international founder. * Optimize your PC for higher RAC. * SETI@home needs your help. * |
petri33 Send message Joined: 6 Jun 02 Posts: 1668 Credit: 623,086,772 RAC: 156 |
Not in linux if I LD_PRELOAD a libarary to override yield() with nanosleep(). To overcome Heisenbergs: "You can't always get what you want / but if you try sometimes you just might find / you get what you need." -- Rolling Stones |
Sutaru Tsureku Send message Joined: 6 Apr 07 Posts: 7105 Credit: 147,663,825 RAC: 5 |
There is also a work around/fix for Windows OSs? And what's with the Windows (XP, Vista, 7, 8) OSs? I ask, I saw PCs with this NV OpenCL BUG (run time = CPU time), and I saw some PCs which don't have him (run time ≠CPU time). Hmm .., maybe they have set 2+ AP WUs/GPU, but just 1 AP GPU was calculated with 1 or 2 additional SAHv7 WUs, so the run time ≠CPU time? Strange. * Best regards! :-) * Philip J. Fry, team seti.international founder. * Optimize your PC for higher RAC. * SETI@home needs your help. * |
Sutaru Tsureku Send message Joined: 6 Apr 07 Posts: 7105 Credit: 147,663,825 RAC: 5 |
Example: Win7 x64, GTX780: 327.23 WHQL driver available Someone have this example combi? You still see the 100 % CPU-thread usage/AP OpenCL NV GPU app (if 2+ AP WUs/GPU simultaneously calculation)? Thanks. * Best regards! :-) * Philip J. Fry, team seti.international founder. * Optimize your PC for higher RAC. * SETI@home needs your help. * |
jason_gee Send message Joined: 24 Nov 06 Posts: 7489 Credit: 91,093,184 RAC: 0 |
Example: Win7 x64, GTX780: 327.23 WHQL driver available Yes you will see 100% usage. That's not a bug. "Living by the wisdom of computer science doesn't sound so bad after all. And unlike most advice, it's backed up by proofs." -- Algorithms to live by: The computer science of human decisions. |
Sutaru Tsureku Send message Joined: 6 Apr 07 Posts: 7105 Credit: 147,663,825 RAC: 5 |
But if I look to ATI/AMD GPUs if they make AP OpenCL WUs, the CPU time is much smaller than the run time. ;-) You know, why on NVidia GPUs this happen? Is the CPU-thread really 100 % used, or maybe he's just 'reserved'? So I can't [EDIT: (don't need to)] hope that the CPU time usage will go down with newer NV drivers? Thanks. * Best regards! :-) * Philip J. Fry, team seti.international founder. * Optimize your PC for higher RAC. * SETI@home needs your help. * |
jason_gee Send message Joined: 24 Nov 06 Posts: 7489 Credit: 91,093,184 RAC: 0 |
But if I look to ATI/AMD GPUs if they make AP OpenCL WUs, the CPU time is much smaller than the run time. ;-) Because the application is coded for OpenCL 1.0 and 1.2, not OpenCL 1.1 OpenCL 1.1 changed the specification, you have to program it differently to achieve low CPU usage. I have examples of OpenCL 1.1 applications that use 0% CPU, this style may be used in future apps, and Cuda is moving that way too. (ends up faster and Less CPU,if done right) [Edit:] The 100% are what's called 'spin-waits' which are 'blocking'. Moving to non-blocking design & sending the thread to sleep (0%) is effective, but difficult design-wise. I'll be doing that for Cuda first in x42. experiments have already been done for that (and make everything faster with even less CPU usage) [Edit2:] So I can't [EDIT: (don't need to)] hope that the CPU time usage will go down with newer NV drivers?It's possible they might put in some hack one day like OpenCl 1.2, but this will always be less efficient than non-blocking (no 100% spin-waits at all), so that's why Cuda will go this way too & a driver hack to fix one application is IMO unlikely & unnecessary. "Living by the wisdom of computer science doesn't sound so bad after all. And unlike most advice, it's backed up by proofs." -- Algorithms to live by: The computer science of human decisions. |
©2024 University of California
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.