Help with GPU utilization factor needed
21 Feb 2021 3:41:57 UTC
Topic 224882
(moderation:
I'm running an AMD FX 9590 with an Nvidia 2080 and two AMD RX 460's and would like advice on whether or not to change the GPU utilization factor. The CPU and Nvidia GPU are running on a custom cooling loop, the RX 460's are air cooled with the stock dual fans.
PS, wasn't sure which forum to post this into, the FAQ didn't specify.
First do you have enough
)
First do you have enough spare cpu threads to support the extra gpu tasks?
What projects are you running? GR gpu tasks are a lot easier to run with minimal cpu support.
GW gpu tasks need one cpu thread per gpu task. They also need up to 4GB of video memory per task.
Do the 460's have 8GB of memory?
Generally if the per task time is less than 2X the single task time, it is beneficial to run more than single tasks on a card. If combined time is greater, then running multiples has less productivity.
A comment by RX 460's uses would be most helpful to you.
I have all projects set to
)
I have all projects set to 75% of the CPU utilization.
Projects might show if you click on my name.
the 460's are 4gb each, so 8gb in total.
Probably just gonna leave it be as the 460's tend to run hot, even with the fans set to max out around 75C
Khataro wrote:I'm running an
)
The first thing you should do is work out why all your tasks are failing. Forget multiple concurrent tasks until single tasks run correctly. Even then, it's quite likely that multiple concurrent tasks will just cause more problems.
Your tasks list shows all GW tasks as having problems. By clicking on the TaskID link of any failed task (like this one), you can get an idea of what the problem is. Notice that in this particular case you can see:-
Exit status: 197 (0x000000C5) EXIT_TIME_LIMIT_EXCEEDEDin the block of information near the top of the full output list.
Many of the failed tasks show the same elapsed time and the same result. It means that the task wasn't making sufficient progress and a project set time limit (eg 15,037 secs - way beyond the expected crunch time) was exceeded, causing BOINC to terminate the crunching at that point. Notice, most tasks that have exactly the same elapsed time. That limit was for your RX 460s. I noticed a single different value (5,674 secs) which was the time limit for the RTX 2080. You would expect the 2080 to be a lot faster than the RX 460 so the lower time limit is not a surprise.
I have no experience trying to run such disparate GPUs in the one host but I wouldn't be surprised if that were at least part of the problem. My advice would be to run just the single 2080 first, and after a fresh full install of drivers and OpenCL libs from nvidia (not Microsoft), see if you can get proper results with just that GPU. If that works well, you could try adding a single RX 460, reload its drivers and then observe what happens. I don't use Windows at all so have no knowledge about the best way to set this all up properly.
While you attempt to get things running correctly, I would suggest you change preferences to run just the gamma-ray pulsar (GRP) search and NOT the GW search. You are likely to have a much easier time getting things to run correctly that way.
Cheers,
Gary.