Occasional bursts of failing BRP4-Cuda tasks
27 Sep 2012 4:32:06 UTC
Topic 196539
(moderation:
Looking at the results page for one of my PC's I've noticed that in the last month it's had two large bursts of BRP4 tasks which failed almost immediately upon starting. Otherwise the computer is completing tasks successfully.
http://einsteinathome.org/host/2565501/tasks&offset=0&show_names=1&state=5&appid=19
Occasional bursts of failing BRP4-Cuda tasks
)
Hi Dan,
here is the cause:
Which apps are running in parallel? Games? How many GPU tasks per GPU?
KR
Michael
Team Linux Users Everywhere
Should it really be a hard
)
Should it really be a hard failure?
I suppose it's not easy to determine if that's a transient problem like a game or other graphics intensive program being used vs a permanent lack of resources.
I have my GPU preferences set to not compute when the system is in use. That might help.
Joe
BRP4's the only regularly
)
BRP4's the only regularly scheduled GPU app I run, and I'm running them one at a time. Collatz and GPUGrid are both set as backups. The latter's only ran once in the last 40 days (the 26th); the former seems to be running at least a few tasks most days for some reason. I've got a bit more than a day's worth of GPU tasks cached; so it's not the normal server unreachable case that's triggering then.
I've been playing Path of Exile most nights. It's not an exceptionally demanding game, I can get >50FPS @2560x1600 with 2xAA and all other settings at max without having to pause GPU tasks.
The tasks might be trying to run more than once before aborting, I checked a few minutes ago and I've got ~25 BRP4-Cuda's that were ran for a few seconds each before being stopped for some reason.
If I remember, I'll keep boinc manager open on one of my side monitors the next time I start POE to see what happens.
Nothing happened yesterday
)
Nothing happened yesterday night; today I had at least two bursts of tasks starting but only running a few seconds then another task starting. One occured at the same time as a game crash; nothing else notable happeend but I ended up with more started WU's when I checked a second time.
I've a simliar problem on Mac
)
I've a simliar problem on Mac OS X after upgrading from Snow Leopard to Mountain Lion.
For Months there was no problem to use the system in a normal way without having gpu tasks aborted.
The Mountain Lion upgrade also requiered a CUDA update to 5.024, but looking at the logs of DanNeely, this might not be the root cause.
I've got two Macs with NVIDIA-GPU, each with 512MB RAM.
A BRP4-CUDA task will need aprox. 204MB and left aprox. 60-80MB free GPU-RAM. And this ist directly after boot and having only Boincmanager with eventlog, system preferences panel, activity monitor and a terminal open.
Using the desktop will decrease the free GPU-RAM and will not increase after closing the newly started applications.
So I 'sit' here and doing a "cat stderr.txt" in the slot directory of the CUDA Task to supervise the free GPU-RAM.
If I do nothing, the free GPU-RAM remains stable.
At the time, the Macs are supposed to do nothing else then Einstein, so it's not a big problem atm..