| Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
 |
World Community Grid Forums
Category: Active Research
Forum: OpenPandemics - COVID-19 Project
Thread: Trouble Getting GPU Tasks |
| No member browsing this thread |
|
Thread Status: Active Total posts in this thread: 15
|
|
| Author |
|
|
dduggan47
Cruncher Joined: Jan 24, 2006 Post Count: 4 Status: Offline Project Badges: 
|
I'm running WCG on 3 machines. One has been successfully getting GPU tasks since they first started coming out in beta. The other two have not gotten any. One of those units does get GPU tasks successfully from PrimeGrid.
All 3 are running Win10 Pro, use the same profile, have NVIDIA GPUs, and are running BOINC Manager 7.16.11 (x64). I know there was discussion of this kind of thing on the beta thread but I didn't find anything that I'd done wrong (or maybe just nothing I understood). Any ideas appreciated. |
||
|
|
ericcui1
Cruncher Joined: Aug 26, 2017 Post Count: 5 Status: Offline Project Badges: 
|
There aren't enough WUs available, only 3400 per hour. One WU only take 2~3 minutes on a modern GPU. So a couple hundred machines get GPU tasks per hour at most. Considering there are so many crunchers out there, the odds aren't great.
----------------------------------------[Edit 1 times, last edit by ericcui1 at Apr 9, 2021 1:09:44 PM] |
||
|
|
ThreadRipper
Veteran Cruncher Sweden Joined: Apr 26, 2007 Post Count: 1317 Status: Offline Project Badges: 
|
Have you tried closing BOINC and then starting it again and immediately checking the event log in BOINC. Make sure it doesn't say "Using missing NVIDIA GPU" or something like that. I've had a couple of machines with that issue. A driver update with the "clean install" option resolved that issue.
----------------------------------------However, right now OPNG WUs are very few so I am also only getting a couple of WUs each day on different machines.  Join The International Team: https://www.worldcommunitygrid.org/team/viewTeamInfo.do?teamId=CK9RP1BKX1 AMD TR2990WX @ PBO, 64GB Quad 3200MHz 14-17-17-17-1T, RX6900XT @ Stock AMD 3800X @ PBO AMD 2700X @ 4GHz |
||
|
|
spRocket
Senior Cruncher Joined: Mar 25, 2020 Post Count: 251 Status: Offline Project Badges: 
|
Another thing to consider is whether or not the GPU is supported. Old Nvidia 9xxx/7xxx cards are out, since they only support OpenCL 1.0 or at best 1.1.
|
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
I've not had any luck since the launch. Using an AMD 5600XT and I've checked in the event log that OpenCL 1.2 is recognized and that the client is requesting GPU workunits. I did receive a couple of beta test units, so the configuration is correct. Hope the availability increases over time.
|
||
|
|
dduggan47
Cruncher Joined: Jan 24, 2006 Post Count: 4 Status: Offline Project Badges:
|
There aren't enough WUs available, only 3400 per hour. One WU only take 2~3 minutes on a modern GPU. So a couple hundred machines get GPU tasks per hour at most. Considering there are so many crunchers out there, the odds aren't great. Thanks for responding. The only thing that makes me doubt that's the problem is that the one device of the three that's getting tasks has been getting them consistently. At the moment one is running and 19 are queued! That's about par for the course when tasks have been available. |
||
|
|
dduggan47
Cruncher Joined: Jan 24, 2006 Post Count: 4 Status: Offline Project Badges:
|
Have you tried closing BOINC and then starting it again and immediately checking the event log in BOINC. Make sure it doesn't say "Using missing NVIDIA GPU" or something like that. I've had a couple of machines with that issue. A driver update with the "clean install" option resolved that issue. However, right now OPNG WUs are very few so I am also only getting a couple of WUs each day on different machines. Thanks threadripper. I tried that on both of the machines in question. On my HP Envy I got the following: 4/9/2021 5:11:20 PM | | CUDA: NVIDIA GPU 0: GeForce GTX 745 (driver version 441.06, CUDA version 10.2, compute capability 5.0, 4096MB, 3427MB available, 793 GFLOPS peak) 4/9/2021 5:11:20 PM | | OpenCL: NVIDIA GPU 0: GeForce GTX 745 (driver version 441.06, device version OpenCL 1.2 CUDA, 4096MB, 3427MB available, 793 GFLOPS peak) To my uneducated eye it looks hunky dory so still not sure what's amiss. The Spectre15 laptop had more interesting results: 4/9/2021 5:21:50 PM | | CUDA: NVIDIA GPU 0: GeForce MX150 (driver version 460.89, CUDA version 11.2, compute capability 6.1, 2048MB, 1662MB available, 1176 GFLOPS peak) 4/9/2021 5:21:50 PM | | App version needs OpenCL but GPU doesn't support it 4/9/2021 5:21:50 PM | Amicable Numbers | Application uses missing NVIDIA GPU 4/9/2021 5:21:50 PM | | app version refers to missing GPU type intel_gpu 4/9/2021 5:21:50 PM | Einstein@Home | Application uses missing intel_gpu GPU 4/9/2021 5:21:50 PM | | app version refers to missing GPU type intel_gpu 4/9/2021 5:21:50 PM | Einstein@Home | Application uses missing intel_gpu GPU 4/9/2021 5:21:50 PM | | App version needs OpenCL but GPU doesn't support it 4/9/2021 5:21:50 PM | Einstein@Home | Application uses missing NVIDIA GPU 4/9/2021 5:21:50 PM | | App version needs OpenCL but GPU doesn't support it 4/9/2021 5:21:50 PM | Milkyway@Home | Application uses missing NVIDIA GPU 4/9/2021 5:21:50 PM | | App version needs OpenCL but GPU doesn't support it 4/9/2021 5:21:50 PM | Moo! Wrapper | Application uses missing NVIDIA GPU 4/9/2021 5:21:50 PM | | App version needs OpenCL but GPU doesn't support it 4/9/2021 5:21:50 PM | NumberFields@home | Application uses missing NVIDIA GPU I'd say that answered the question except that there's no error message there from WCG. |
||
|
|
Richard Haselgrove
Senior Cruncher United Kingdom Joined: Feb 19, 2021 Post Count: 360 Status: Offline Project Badges:
|
There aren't enough WUs available, only 3400 per hour. One WU only take 2~3 minutes on a modern GPU. So a couple hundred machines get GPU tasks per hour at most. Considering there are so many crunchers out there, the odds aren't great. Thanks for responding. The only thing that makes me doubt that's the problem is that the one device of the three that's getting tasks has been getting them consistently. At the moment one is running and 19 are queued! That's about par for the course when tasks have been available. One other thing that may be in play is the way BOINC handles shortages in work. Once a machine has started getting work, it keeps on asking - and it's more likely to get more work, if it asks at the right moment. On the other hand, if BOINC doesn't get any work, it assumes it's not worth asking, and waits longer and longer between attempts - thus making it even less likely to be successful on the rare attempts it does make. That's intended to protect the project's server if its under-powered and struggling to keep up. But it isn't very helpful in situations such as we have now, where the few successful computers tend to go on being lucky. An alternative would be for the project team to set the server to request that clients contact it at regular minimum intervals - every one or two hours, say. They have that setting available to them. It would mean that the client user-base would be much quicker to respond if work became available from the researchers in larger quantities. |
||
|
|
widdershins
Veteran Cruncher Scotland Joined: Apr 30, 2007 Post Count: 674 Status: Offline Project Badges: 
|
Filling up on short WU's from CPU WCG projects and then switching profile to a GPU only profile seems to net a few when the CPU projects report in and ask for new work regularly. Helps to be running a machine processing 24 cpu units at a time though.
 |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
I just received the first GPU work unit. Glad it's working as it should, got a bit worried for a while there.
|
||
|
|
|