Checkpoint file missing - Starting from scratch...
4 Dec 2016 17:18:36 UTC
Topic 203439
(moderation:
Hi all,
after not being able to get my R9 280x to crunch under Mac OSX, I've reverted to a GTX 580 on this machine. I was just browsing through some of the output, and found this...
[16:57:31][27127][INFO ] Using CUDA device #0 "GeForce GTX 580" (512 CUDA cores / 1581.06 GFLOPS) [16:57:31][27127][INFO ] Version of installed CUDA driver: 8000 [16:57:31][27127][INFO ] Version of CUDA driver API used: 3020 [16:57:31][27127][INFO ] Checkpoint file unavailable: p2030.20161002.G58.46+00.21.N.b5s0g0.00000_720.cpt (No such file or directory). ------> Starting from scratch... [16:57:31][27127][INFO ] Header contents: ------> Original WAPP file: ./p2030.20161002.G58.46+00.21.N.b5s0g0.00000_DM72.00
Is this normal? It's giving me a runtime of 1198 (s) and CPU time of 164 (s) ... Is that ok? Or should the runtime be only marginally larger than the CPU time? See the whole output here: https://einsteinathome.org/task/592949698
Thanks for any help - I'm a beginner with all this!
Bye for now,
Kailee71.
I believe those runtimes are
)
I believe those runtimes are normal. I looked at my own results with Nvidia and Windows and there are same kind of results (runtime roughly ten times the CPU time).
Checkpoint file unavailable and Starting from scratch is normal when a new task starts and it haven't been running long enough earlier to have recorded any checkpoint yet. You can try and see what the log says after you let a task run for some time and then suspend computing and exit Boinc completely and then start it again and resume computing.
To add to what Richie said
)
To add to what Richie said the BRP4G tasks are actually 16 BRP4 tasks bundled together, so the stderr will log start up and completion messages for 16 subtasks. If you read the log carefully you will see that the app has the time to checkpoint 1 time per subtask before it's done and a new one starts up. If you were to shut down in the middle of one of these tasks you should get messages about how many subtasks are already completed before the current one resumes.
As to run time vs CPU time it's normal and desired for a GPU task to have longer run time than CPU time. The CPU time reported is the amount of the your CPU spent supporting the GPU app, this should be as low as possible while the run time is the total wall clock time the task took to complete.
Kai Leibrandt wrote:
)
To add further to what others have said, lines in these logs are marked with a tag (eg. [INFO] in this case) which tells you what the message is about. The only time you need to be concerned is if the tag is [ERROR] because that means there really was a problem. [INFO] lines are just the normal chatter expected in a log file. These files are logs only and are not the scientific results being returned. There are separate small 'result' files uploaded, one for each of the 16 sub-tasks in a bundled task. If you watch the transfers tab of BOINC Manager, you can briefly see them being uploaded just after a task completes.
Cheers,
Gary.