cuda erroring out on every client
25 Dec 2009 11:46:34 UTC
Topic 194685
(moderation:
All the errors look the same. Lots of debug info but makes no sense to me. Maybe someone can figure it out. In mean time I am running collatz and seti cuda just fine.
http://einsteinathome.org/host/2034463/tasks
windows 7 amd system 64 bit, 9800gtx+ rarely used for anything other than boinc.
thanks for looking
cuda erroring out on every client
)
BOINC 6.10.11 was a development version, not stable. Perhaps you want to try updating the client at least to something of the latest.
Other than that, your error is "Maximum elapsed time exceeded". This means that your GPU has taken more floating point operations at that point than are given to the task through the n value. You can try to edit this number (in client_state.xml) on any unstarted task, increase it with a nought, see if that stabilizes it.
Warning! If you do plan on editing client_state.xml to edit this value, make sure you exited BOINC fully and that you do not edit anything else in the file. Any changes in the file that aren't correct can cause complete loss of work for all projects.
Other than that, I see you're running the 195.62 drivers. These have been reported as being problematic on some systems. You may want to try a previous driver version first.
RE: All the errors look the
)
Whilst Jord has given you the answer about what you need to do to stop the errors, it's a shame you weren't paying attention when the problem was first identified and solutions given.
You are still using version 3.10 of the CUDA app. This is the beta test version which Bernd announced here as being replaced by an official version. He asked that all beta testers should stop using the old version and move to the official one. The current official version is 3.13, which will download automatically if you remove your app_info.xml.
The problem you are having was first mentioned nearly three months ago in the 3.10 CUDA app beta test thread. If you read the linked post and the messages that follow you will see that the problem is identified as one that was introduced in a particular alpha version of BOINC - no doubt the one you are still using. A couple of days later, as reported in this message from Richard Haselgrove, a fix was checked into BOINC which I believe did fix the problem as there haven't been any further reports since that time. The fixed version was (I believe) 6.10.13 or perhaps the one after that. You should really be using the released version 6.10.18. If you upgrade to this recommended version, you should not need to do the editing that Jord mentions.
If you insist on using alpha versions of BOINC, you should be paying attention and upgrading as soon as later versions are released, particularly later versions that are reported to fix known problems. If you are too busy to pay close attention, you really shouldn't be using alpha versions. You just waste other people's time when reporting problems that were fixed such a long time ago, not to mention the waste of electricity for all the failed results.
Cheers,
Gary.