Look at your app_version listing from last time. Einstein distributes CUDA DLLs with additional version information in the names, and renames them on copying to the slot directory. That bit's OK.
Thanks, Richard. Looking at client_state.xml it appears that the two file references I described as not being found in the Windows directory are renames for these two files, which I do find in F:\Program Data\BOINC\projects\einstein.phys.uwm.edu (which is where I've got the BOINC Einstein stuff on my system)
07/03/2015 06:48 AM 298,272 cudart_xp64_55_22.dll
07/03/2015 06:48 AM 74,730,784 cufft_xp64_55_22.dll
The July 3 timestamps imply I got these two files when downloading the previous attempt at a Windows CUDA55, and still have them.
Thanks for correcting my mistaken attempt to identify something possibly missing.
I have a full set of eleven 1.56 tasks running on five GPUs on three hosts, GTX 660s, 750, 750Ti, and 970. With 17 to 32 minutes of run time, all currently appear to be progressing normally.
I did think I saw unexpectedly high CPU consumption during the first minute of startup on at least one task.
I don't see any substantial increase in GPU temperatures, so I don't think there has been a really large increase in power consumption. My 970 is heavily memory overclocked, and moderately GPU overclocked, so had not much margin before, so I am encouraged it has not yet crashed with what might be expected to be a possibly higher challenge from the new code.
Of my initial batch of eleven, seven have run to completion, and the other four are making apparently good progress.
One task has already validated which is comforting, as the quorum partners of the others look unlikely to reply for some time. My thanks to stoneageman for running a short task queue.
An extremely rough initial view of performance is a definite improvement, perhaps somewhere in the 20% productivity improvement range. Power seems little changed, though it may be up moderately, not a lot.
Regarding CPU consumption, I do think I saw an initially higher than expected spike for perhaps a minute at the start of at least one task, but that did not persist, and the total CPU resource consumption to complete a result for the seven completed work units is modestly down from the previous application.
RE: Look at your
)
Thanks, Richard. Looking at client_state.xml it appears that the two file references I described as not being found in the Windows directory are renames for these two files, which I do find in F:\Program Data\BOINC\projects\einstein.phys.uwm.edu (which is where I've got the BOINC Einstein stuff on my system)
07/03/2015 06:48 AM 298,272 cudart_xp64_55_22.dll
07/03/2015 06:48 AM 74,730,784 cufft_xp64_55_22.dll
The July 3 timestamps imply I got these two files when downloading the previous attempt at a Windows CUDA55, and still have them.
Thanks for correcting my mistaken attempt to identify something possibly missing.
RE: New version will be
)
1.56 is out.
BM
BM
Sorry, but still not working
)
Sorry, but still not working for me (Computing Error)
Arthur
I know I am a part of a story that starts long before I can remember and continues long beyond when anyone will remember me [Danny Hillis, Long Now]
RE: 1.56 is out. Still
)
Still shows the same symptom (illegal instruction).
Retracted for now.
BM
BM
The 64Bit Windows CUDA build
)
The 64Bit Windows CUDA build doesn't work, trying the 32Bit one. Published new Beta app version.
BM
BM
Seems to be working, promoted
)
Seems to be working, promoted one tasks to run and it has now reached the first checkpoint. Now waiting for it to finish and validate!
Forced a few v1.56 to run on
)
Forced a few v1.56 to run on XP32/GTX 750 Ti. So far so good, will keep watching...
-----
I have a full set of eleven
)
I have a full set of eleven 1.56 tasks running on five GPUs on three hosts, GTX 660s, 750, 750Ti, and 970. With 17 to 32 minutes of run time, all currently appear to be progressing normally.
I did think I saw unexpectedly high CPU consumption during the first minute of startup on at least one task.
I don't see any substantial increase in GPU temperatures, so I don't think there has been a really large increase in power consumption. My 970 is heavily memory overclocked, and moderately GPU overclocked, so had not much margin before, so I am encouraged it has not yet crashed with what might be expected to be a possibly higher challenge from the new code.
RE: I did think I saw
)
I saw this too, but now (after almost 1 hour) the CPU usage is down to ~12%.
-----
Of my initial batch of
)
Of my initial batch of eleven, seven have run to completion, and the other four are making apparently good progress.
One task has already validated which is comforting, as the quorum partners of the others look unlikely to reply for some time. My thanks to stoneageman for running a short task queue.
An extremely rough initial view of performance is a definite improvement, perhaps somewhere in the 20% productivity improvement range. Power seems little changed, though it may be up moderately, not a lot.
Regarding CPU consumption, I do think I saw an initially higher than expected spike for perhaps a minute at the start of at least one task, but that did not persist, and the total CPU resource consumption to complete a result for the seven completed work units is modestly down from the previous application.