| Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
 |
World Community Grid Forums
Category: Active Research
Forum: OpenPandemics - COVID-19 Project
Thread: OpenPandemics - GPU Stress Test |
| No member browsing this thread |
|
Thread Status: Active Total posts in this thread: 781
|
|
| Author |
|
|
Jorlin
Advanced Cruncher Deutschland Joined: Jan 22, 2020 Post Count: 90 Status: Offline Project Badges: 
|
Does the "Allow research to run on my CPU?" profile setting consider an integrated GPU to be a CPU?  If so, I suppose that would make sense, but it wasn't the behavior that I expected. I think it's because the GPU threads also claim part of a CPU (as in the job description), which they wouldn't be allowed to use if CPU computing is disabled. Hmm... never tried to set the CPU to zero in the app_config.xml ...  |
||
|
|
William Albert
Cruncher Joined: Apr 5, 2020 Post Count: 41 Status: Offline Project Badges: 
|
I think it's because the GPU threads also claim part of a CPU (as in the job description), which they wouldn't be allowed to use if CPU computing is disabled. One of my machines has an old AMD CPU (without any integrated graphics) and a modern Nvidia GPU. It's assigned a profile where CPU work is explicitly disabled, and that grabs and crunches Nvidia GPU WUs without any problems. |
||
|
|
Jorlin
Advanced Cruncher Deutschland Joined: Jan 22, 2020 Post Count: 90 Status: Offline Project Badges:
|
I think it's because the GPU threads also claim part of a CPU (as in the job description), which they wouldn't be allowed to use if CPU computing is disabled. One of my machines has an old AMD CPU (without any integrated graphics) and a modern Nvidia GPU. It's assigned a profile where CPU work is explicitly disabled, and that grabs and crunches Nvidia GPU WUs without any problems. so much for my theory  |
||
|
|
Grumpy Swede
Master Cruncher Svíþjóð Joined: Apr 10, 2020 Post Count: 2511 Status: Recently Active Project Badges:
|
I have an Intel iGPU HD4600, and CPU disabled in my Profile. I get iGPU tasks without any problems.
|
||
|
|
Ian-n-Steve C.
Senior Cruncher United States Joined: May 15, 2020 Post Count: 180 Status: Offline Project Badges: 
|
Does the "Allow research to run on my CPU?" profile setting consider an integrated GPU to be a CPU? If so, I suppose that would make sense, but it wasn't the behavior that I expected. I think it's because the GPU threads also claim part of a CPU (as in the job description), which they wouldn't be allowed to use if CPU computing is disabled. Hmm... never tried to set the CPU to zero in the app_config.xml ... that's not how it works. I have CPU processing disabled and I still get GPU work which is requesting 1CPU core per GPU. when you click the project side check box for CPU work, that has to do with the types of WUs sent to you, it's not so strict that it means no CPU can be used as all. just no CPU WUs. any GPU app will need *SOME* CPU support in some way. different devices and/or applications require different amounts of support.  EPYC 7V12 / [5] RTX A4000 EPYC 7B12 / [5] RTX 3080Ti + [2] RTX 2080Ti EPYC 7B12 / [6] RTX 3070Ti + [2] RTX 3060 [2] EPYC 7642 / [2] RTX 2080Ti |
||
|
|
Jorlin
Advanced Cruncher Deutschland Joined: Jan 22, 2020 Post Count: 90 Status: Offline Project Badges:
|
I already called my theory debunked ;)
---------------------------------------- |
||
|
|
JKarhu
Cruncher Joined: Apr 5, 2021 Post Count: 5 Status: Offline |
Sorry right away as I am sure this has been discussed, I just can't find the info from the swamp:
All my GPU tasks fail with "computation error". Logs say "GPU not found: type=ATI, opencl_device_index=1, device_num=0". I am running on AMD eGPU, on Mac. Any advice? Thank you in advance! |
||
|
|
goben_2003
Advanced Cruncher Joined: Jun 16, 2006 Post Count: 146 Status: Offline Project Badges: 
|
Thanks, -Uplinger Hi, I still have not get a single GPU work unit for Intel HD520 and Intel UHD GPUs (been running for couple of weeks now) . Intel HD 4600 is getting some. 3 days ago I reinstalled the win10 on HD 520 laptop, but that didn't seem to help either. (trimmed) Which intel UHD GPUs? I am running on 2 UHD 620 and 1 UHD 630. Where was it installed from? One of my machines was the victim of Windows 10 auto upgrading the driver. I downloaded the full driver (not from microsoft) and reinstalled, then it worked. Not sure if it the driver version actually matters, but here are the ones I am using: 1 UHD with 26.20.100.7262, 1 UHD 620 with 2620.100.8142, 1 UHD 630 with 27.20.100.8681. Rolled back drivers to 26.20.100.xxxx and now boinc shows CL2.1, but the problem persists. I think I'll install linux on one these machines to see if it's better Sorry that did not work. I have been thinking about this issue of requesting work, it is available, but not receiving any. So I tried different things trying to reproduce the issue. I was able to reproduce it by turning off "Allow research to run on my CPU?" on the device profile on the WCG website. This was with work unit cache settings and OPN project limit set high enough to not be a limiting factor. The log showed it requesting intel gpu work units, but not getting any. Then I enabled "Allow research to run on my CPU?" and started receiving intel gpu work units again. I repeated this and got the same results. I saw in a previous post of yours that you have set the profile to not use the CPU. Just as a test, have you tried enabling CPU usage? You can set the project limit OPN to 1 so you do not get many cpu work units while testing this, but still allowing 1 opn gpu unit to be sent. Hopefully that works.  |
||
|
|
flensr
Cruncher Joined: Oct 31, 2018 Post Count: 25 Status: Offline Project Badges: 
|
I don't know if it counts as a bug or not, but one of my 2 computers wasn't able to start participating initially because of some other WCG results or download that was stalled for about a day. Because that one non-GPU WU transfer was stalled, WCG was refusing to send new work. I had to manually delete the stalled transfer and stop/restart BOINC. After that, it has worked fine.
---------------------------------------- |
||
|
|
spRocket
Senior Cruncher Joined: Mar 25, 2020 Post Count: 280 Status: Offline Project Badges: 
|
A quick check on my "instantaneous" stats in boinctui, an hour before the UTC day ends, and I'm doing better than yesterday, with time for a wingman to finish up a few units awaiting one. I'm going to leave my two-GPU-tasks configuration going until the stress test ends, as I can already observe from the queue that most of the time, if one GPU task is in its preliminary CPU stage, the other one isn't. It's like getting an extra fraction of a GPU for the cost of a CPU slot.
----------------------------------------Once we're back to reality and GPU work units become scarce again, I'll switch back to 15 CPU+1 GPU. No sense tying up a CPU slot in that situation. ETA: On my Ryzen 7 1700, the CPU stage of these big GPU work units typically lasts about four minutes or so, out of typically 9-13 minutes total run time with a single GPU task; that's a pretty good chunk of time that the GPU isn't being used. With two GPU tasks, the tasks now take almost twice as long (surprise!), but that gap in GPU usage generally gets filled with two tasks running, a net win. [Edit 1 times, last edit by spRocket at Apr 28, 2021 11:30:56 PM] |
||
|
|
|