I've found this error with the unchecked "Use GPU while computer is in use" checkbox.
If you don't use the computer, the progress is running correctly, but if you start working, the progress stops (that is correct !!) but if you again stop working, the progress restart from the beginning (0.0)
But the worst is that if you repeat this behaviour more times, the elapsed time can reach many hours, and the WU aborts with a "Computation error".
WinXP 64b - BM 6.10.16 64b - GTX260 - CUDA 191.07
____________
/ 
Reply
Quote
I've found this error with the unchecked "Use GPU while computer is in use" checkbox.
If you don't use the computer, the progress is running correctly, but if you start working, the progress stops (that is correct !!) but if you again stop working, the progress restart from the beginning (0.0)
But the worst is that if you repeat this behaviour more times, the elapsed time can reach many hours, and the WU aborts with a "Computation error".
WinXP 64b - BM 6.10.16 64b - GTX260 - CUDA 191.07
Activating the option "leave application in memory" should fix that.
/ Reply
Quote
Sorry, but the option "leave application in memory" was already activated.
And this problem is specific only to this project, all the other CUDA projects are working fine in this situation
____________
/ Reply
Quote
I can confirm this "feature" (with BOINC 6.10.11 and 6.10.16)
WinXP32, Quadro fx3800 , CUDA 190.38
/ Reply
Quote
Try Cuda 190.62
I have both GTX 260's and GTX 295's installed. Generally I only get this problem when quitting and then restarting BOINC.
Running Vista, Suspend to memory and I also use GPU while computer in use.
EDIT: I also use 6.10.13, 6.10.14 and 6.10.16.
____________
/ Reply
Quote
@David.
Warning! I've said "Not" to use GPU while computer is in use in the Boinc Manager Advanced Preferences, whilst in the Milkyway Computer preferences there is "Use GPU while computer is in use"
BTW I submitted this problem to BOINC Alpha Test
____________
/ Reply
Quote
Sorry, but the option "leave application in memory" was already activated.
And this problem is specific only to this project, all the other CUDA projects are working fine in this situation
I remember there was a discussion that this setting only applies to CPU applications and not GPU applications. GPU applications may never stay in memory as there was the fear that this could disturb other applications using the graphics RAM. That means the MW GPU applications restart from the last checkpoint.
So either what you are seeing is that the CUDA app doesn't checkpoint (the ATI app doesn't either) or you have set the time between checkpoints ("write to disk at most every xxx seconds") to a high value. That means the app doesn't checkpoint because not enough time has passed until the execution is interrupted.
/ Reply
Quote
@cluster physik.
I've never changed the "Write to disk at most every" parameter in all the projects I'm crunching, as far as I remember
BTW this parameter, in Milkyway, is 60 seconds
/ Reply
Quote
Despite the new 0.24 version, the problem I sumbitted has not been resolved !!!!
/ Reply
Quote
Now the only way to bypass this problem is, when I receive some WU, to put the client in "Use GPU while computer is in use", wait untill all the WUs terminate and then restore the previous situation with "Don't use GPU while computer is in use".
/ Reply
Quote
I waited some news about the problem I submitted, but nothing happened !!
This problem is still alive!! also in the .24 (cuda 23) version
Is there an estimated of a solution ??
/ Reply
Quote
Milkyway has no checkpoints, and i doubt it will ever have any, so no, if you stop a Milkyway WU, it restart from 0
But a modern GPU crunch a WU in few minutes, so it isn't really a problem...
/ Reply
Quote