Hello everyone,

I'm running Primegrid (PPS Sieve) on a nvidia Geforce 8400 GS only (no cpu). I know that it doesn't have a lot of power, but I had it lying around, and had a rig without a GPU yet, so I popped it in.

Unfortunately, I always get lots of tasks, and my GPU can't finish them all before the deadline. It takes about 20 hours to complete a task with this GPU. When BOINC requests new tasks it will get a batch of about 20-30 WUs with a deadline of 6 days. At first the estimated run time was only 1 hour per WU, so I waited until it adjusted to a realistic estimate and aborted the tasks that wouldn't have finished in time. I though that this would only be a one-time occurence and would now be fixed with the new estimates.

But now, I got a large batch of tasks again (about 20), even though BOINC should know better now (runtime estimates are about 20 hours which is realistic). I could handle about 6 or 7 before the deadline, but not 20.

There is no way I can finish them. And because BOINC notices this, the WUs run on high priority and BOINC will waste a CPU (which would go to another project I crunch on) by reserving it for the GPU task alone, leading to lots of CPU idle time.

Is there anything I can do? Do I harm the project if ~90% of my WUs time out?

Thanks!
David

I have the same issue using an old ATI board on the PPS Sieve. I set the DCF to 1 originally and the tasks stated about 8.5 minutes to complete. Naturally, this old board takes about 25 times that long so the DCF has corrected to that. However, the client always downloads 30 or so tasks and with only several days to complete these tasks, it goes into high priority. I just have to calculate how many it can really complete and abort the rest. Does it every time even with the DCF adjusted, so once it completes the last task (usually in high priority), it then downloads another large batch. So probably just have to get used to it.

On a side note, I haven't noticed a CPU being wasted with a GPU task in high priority. The PPS task isn't taking a full CPU in this state or even close. Maybe it's an NVIDIA thing, though I'm running an ATI under Linux, so maybe other variables in play.

I have tried to let the WUs time out (that's why it took so long to report back). I tried aborting the tasks before, but that didn't work.

Unfortunately, I just got 71 (!) new tasks. That's more than 60 days of work for my system!

Only one of my two cores is utilized (one GPU task running in high priority, one CPU task from another project running, one CPU task waiting to run). My personal priority lies on the other project and Primegrid only runs to fill the otherwise unused GPU, so I would really like to get BOINC to use both CPU cores again.

Any ideas?

I have tried to let the WUs time out (that's why it took so long to report back). I tried aborting the tasks before, but that didn't work.

Unfortunately, I just got 71 (!) new tasks. That's more than 60 days of work for my system!

Only one of my two cores is utilized (one GPU task running in high priority, one CPU task from another project running, one CPU task waiting to run). My personal priority lies on the other project and Primegrid only runs to fill the otherwise unused GPU, so I would really like to get BOINC to use both CPU cores again.

Any ideas?

Thanks for your input, I tried something different again.

The "Waiting to run" wasn't because of the high priority of Primegrid, but only because my machine needed more RAM. Tweaking the memory settings helped already, now all three tasks are running, and I will try to get a bit more physical RAM for my machine.

Only one question remains: Would I harm the project if I just let all these WUs time out and only finish some of them? I don't want to check this box every day :)

If it's fine I'll just let it run like this. If it's not I will try the suggestion in the last reply.

Thanks for your input, I tried something different again.

The "Waiting to run" wasn't because of the high priority of Primegrid, but only because my machine needed more RAM. Tweaking the memory settings helped already, now all three tasks are running, and I will try to get a bit more physical RAM for my machine.

Only one question remains: Would I harm the project if I just let all these WUs time out and only finish some of them? I don't want to check this box every day :)

If it's fine I'll just let it run like this. If it's not I will try the suggestion in the last reply.

That's what I wanted to know. I'll just let it run unsupervised now.

Thank you.

That's what I wanted to know. I'll just let it run unsupervised now.

Thank you.

I'm also getting a ridiculous amount of PrimeGrid work units lately and have tried everything to change that, but nothing works. And it is doing so on all of my client machines. I've let work units time out, I've aborted them multiple times in a row, I've reset the project a dozen times, I've even removed and reinstalled the project. Even changing my account preferences to NO GPU work does not stop these downloads of an impossible amount of GPU work!

The PG scheduler is still sending me forty or more 12 hour projects for my single old ATI GPU that have a deadline within the next two days. Obviously, the problem is on the PrimeGrid scheduler end. The scheduler is no longer paying any attention to preferences or benchmarks or common sense. The main problem with all of this is that due to the time deadlines, the client is prioritizing all the PrimeGrid work and none of my other projects are getting work done.
So I will simply suspend the PrimeGrid account until there is notification that this has been repaired.

I'm also getting a ridiculous amount of PrimeGrid work units lately and have tried everything to change that, but nothing works. And it is doing so on all of my client machines. I've let work units time out, I've aborted them multiple times in a row, I've reset the project a dozen times, I've even removed and reinstalled the project. Even changing my account preferences to NO GPU work does not stop these downloads of an impossible amount of GPU work!

The PG scheduler is still sending me forty or more 12 hour projects for my single old ATI GPU that have a deadline within the next two days. Obviously, the problem is on the PrimeGrid scheduler end. The scheduler is no longer paying any attention to preferences or benchmarks or common sense. The main problem with all of this is that due to the time deadlines, the client is prioritizing all the PrimeGrid work and none of my other projects are getting work done.
So I will simply suspend the PrimeGrid account until there is notification that this has been repaired.

I just loaded windows 7 to do some testing and I'm getting the same results.
I have my buffers set to 0.00 Max and 0.00 Min.
It just never stops downloading. Very strange. I've not seen this before. I know it does not happen on Linux.

Edit: I was running the beta of boinc 7.2.5. Downgraded to recommended version and it stopped downloading.

I just loaded windows 7 to do some testing and I'm getting the same results.
I have my buffers set to 0.00 Max and 0.00 Min.
It just never stops downloading. Very strange. I've not seen this before. I know it does not happen on Linux.

Edit: I was running the beta of boinc 7.2.5. Downgraded to recommended version and it stopped downloading.