It would be especially interesting to test wrt. reports of severe underperformance of newer NVIDIA Linux drivers, e.g. see here: http://einsteinathome.org/node/197093.
I was running 319 something at the time of BRP5 but reversed back to old faithful 310.32 - mainly because i needed to understand what was causing errors , i don´t recall 319 being particularly slow. see http://einsteinathome.org/node/197035
The weekend has started, and so has 325.15
Fri 09 Aug 2013 22:18:32 BST NVIDIA GPU 0: GeForce GTX 460 (driver version unknown, CUDA version 5050, compute capability 2.1, 768MB, 163 GFLOPS peak)
Fri 09 Aug 2013 22:18:32 BST NVIDIA GPU 1: GeForce GTX 460 (driver version unknown, CUDA version 5050, compute capability 2.1, 768MB, 163 GFLOPS peak)
As always, forgot to wait for boinc to go down before working in a std ctrl-alt-f1 terminal, which generated a number of errored WUs. (this was the error issue referred above)
These are basically me being careless in 310.32, not the upgrade to 325.15 specifically.
Current profile is E@H only - BRP4 (cpu) and BRP5 (gpu) which is not gpu optimal. Running only 2 of 4 cores on a i3, and at most 50% cpu.
Previous 310.32 results BRP5
Elapsed times per BRP5 WU (2 tasks/gpu)
gpu0 18000s (+/-500s)
gpu1 39000s (+/-1000s)
Running a gpu task only profile last week the gp0 figures were about
gpu0 16500s (+/-500s)
gpu1 38500s (+/-1000s)
Will post new results in a few hours, as the first fully baked at 325.15 are starting to rise. No video issues or smoke as yet.
* The first completed gpu1 tasks started life on gpu0 in 310.32, so their times - 38252s, 38346s - are probably lower than the would be in normal running, as their first ~800s were spent on gpu0.
I'm successfully running the new "short lived branch" driver 325.15 for a few days now, no problems and no change in run time.
GTX 570 on kernel 3.10.4 (distro Fedora 18 64-bit), BRP5 takes 14960s per WU on average (running 3 WU simult.).
nVidia 325.15 Linux 64 bit released Aug 5
)
Hi!
It would be especially interesting to test wrt. reports of severe underperformance of newer NVIDIA Linux drivers, e.g. see here: http://einsteinathome.org/node/197093.
Cheers
HB
HB hi I was running 319
)
HB hi
I was running 319 something at the time of BRP5 but reversed back to old faithful 310.32 - mainly because i needed to understand what was causing errors , i don´t recall 319 being particularly slow. see http://einsteinathome.org/node/197035
The weekend has started, and so has 325.15
Fri 09 Aug 2013 22:18:32 BST NVIDIA GPU 0: GeForce GTX 460 (driver version unknown, CUDA version 5050, compute capability 2.1, 768MB, 163 GFLOPS peak)
Fri 09 Aug 2013 22:18:32 BST NVIDIA GPU 1: GeForce GTX 460 (driver version unknown, CUDA version 5050, compute capability 2.1, 768MB, 163 GFLOPS peak)
As always, forgot to wait for boinc to go down before working in a std ctrl-alt-f1 terminal, which generated a number of errored WUs. (this was the error issue referred above)
These are basically me being careless in 310.32, not the upgrade to 325.15 specifically.
Current profile is E@H only - BRP4 (cpu) and BRP5 (gpu) which is not gpu optimal. Running only 2 of 4 cores on a i3, and at most 50% cpu.
Previous 310.32 results BRP5
Elapsed times per BRP5 WU (2 tasks/gpu)
gpu0 18000s (+/-500s)
gpu1 39000s (+/-1000s)
Running a gpu task only profile last week the gp0 figures were about
gpu0 16500s (+/-500s)
gpu1 38500s (+/-1000s)
Will post new results in a few hours, as the first fully baked at 325.15 are starting to rise. No video issues or smoke as yet.
OK maybe some small
)
OK maybe some small improvement.
gpu0(325.15) - 17117s, 17035s, 17361s
gpu1(310.32+325.25) - 38252s, 38346s *
* The first completed gpu1 tasks started life on gpu0 in 310.32, so their times - 38252s, 38346s - are probably lower than the would be in normal running, as their first ~800s were spent on gpu0.
So no problems so far.
I'm successfully running the
)
I'm successfully running the new "short lived branch" driver 325.15 for a few days now, no problems and no change in run time.
GTX 570 on kernel 3.10.4 (distro Fedora 18 64-bit), BRP5 takes 14960s per WU on average (running 3 WU simult.).