|
|
Post by 888 on Jul 7, 2021 22:07:05 GMT
The GPUgrid team have just upgraded their app to use CUDA11.2. Which finally should mean compatibility with nVidia RTX30xx cards.
Unfortunately there are some problems......
1. It appears to be Linux only right now.
2. The app doesn't contain libboost1.74, which means that without manually installing libboost in a particular way, GPUgrid will fail after 2 seconds  I've tried and I cant get it to work. I'd probably have to reinstall graphic card drivers and tweak my system - which I'm not prepared to do for 1 project.
Hopefully they'll fix the app soon with a new version that includes what is needed.
3. There are virtually no WU's anyway.
With any luck they'll sort this out soon. It will be interesting to see what the Ampere cards can do.
|
|
|
|
Post by 888 on Sept 17, 2021 19:52:56 GMT
It looks as though GPUgrid have sorted the problems above, and the other problem they had with multiple GPU systems crunching multiple units only on one GPU.
I think they have Windows versions available too.
I have hit another issue that no-one has reported except me - so it might just be my system, but I dont think so......
Today 800+ of these new units became available and my machine downloaded 4 of them.
2 units were running concurrently on my 2070 and 3070, and all seemed to be fine.
But I checked the progress 1 hour ago and one of the units had failed.
When I checked the logs, it had failed at the exact time that my machine had downloaded and started to run a single WCG grid GPU unit (which are quite rare now - I'm lucky to get 6 a day).
I looks as though this new version doesn't play well with other BOINC projects. So I've disables WCG for now, until I get some answers from GPUgrid.
|
|
|
|
Post by 888 on Sept 20, 2021 18:58:23 GMT
Apparently this problem is 2 years old.
It seems that when a GPUgrid BOINC process is halted, for whatever reason - another BOINC project taking over the resources, a machine reboot, or even pausing the process - then if you are running two or more different models of nvidia GPU, it isn't guaranteed that the process will restart on the same GPU. Resulting in a failed work unit.
I cant believe that they haven't fixed this basic fault in their system after all of this time.
|
|
|
|
Post by Bozo on Sept 22, 2021 22:40:56 GMT
That does seem rather shortsighted. I wonder if that is why I had regular failures when I was running it, not switching gpus but something else temporarily taken the resources from the one I was running?
I don't have the patience to go through logs. something either works or or doesn't
|
|
|
|
Post by Bozo on Dec 2, 2021 23:43:40 GMT
I turned GPUgrid on every task fails after 6 secs updated drivers made no difference currently have around 180 tasks all failed (remember now why i stopped this in first place). going to stop it again for the time being - will have a read when i have more time  |
|
|
|
Post by 888 on Dec 5, 2021 22:46:37 GMT
Yea, GPUgrid is so frustrating. The technical admin side of the project is very poor, and doesn't appear to be improving.
I've ran it for a while now and the frequency of the problems, and time taken to fix issues which shouldn't happen in the first place have made the project unworkable.
Last time I looked, Linux issues were finally settling down, but there still appear to be some problems for some users running windows.....but most weeks the tech admins seem to introduce all new problems, together with reintroduced old ones.
So I've stopped crunching GPUgrid as well, at least until they sort out these technical problems.
|
|
