| Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
 |
World Community Grid Forums
Category: Support
Forum: GPU Support Forum
Thread: GPU work not being sent to me |
| No member browsing this thread |
|
Thread Status: Active Total posts in this thread: 10
|
|
| Author |
|
|
Rizer579
Cruncher Joined: Jun 20, 2020 Post Count: 2 Status: Offline |
I can't seem to receive GPU work for the covid project, I've a one year old good GPU, I've configured in settings for BOINC all the things I can think of to receive GPU work but nothing comes through. What am I doing wrong? I've a Nvidia GeForce 2080 Super GPU, 8GB RAM and all drivers up to date. Thanks.
|
||
|
|
bikeaddict
Cruncher Joined: Apr 11, 2020 Post Count: 32 Status: Offline Project Badges: 
|
First, confirm that BOINC is requesting GPU tasks. In the BOINC Manager app, go to Tools/Event Log... and look for lines that look like this:
Sun 08 Aug 2021 05:30:12 PM CDT | World Community Grid | Requesting new tasks for NVIDIA GPU Sun 08 Aug 2021 05:30:13 PM CDT | World Community Grid | Scheduler request completed: got 0 new tasks Sun 08 Aug 2021 05:30:13 PM CDT | World Community Grid | No tasks sent Sun 08 Aug 2021 05:30:13 PM CDT | World Community Grid | No tasks are available for OpenPandemics - COVID-19 - GPU Next, check your result history. At the top right on the WCG website (where this forum is located) while logged in, go to My Contribution, then Results Status (https://www.worldcommunitygrid.org/ms/viewBoincResults.do) and filter by project and device. Then look for rows starting with OPNG_ to see if you have actually run GPU tasks, but maybe they ran so fast, you didn't see them. GPU tasks are few and far between and would complete very quickly on a 2080 Super, so you may rarely see them in progress. |
||
|
|
william_az
Cruncher Joined: Dec 10, 2008 Post Count: 3 Status: Offline Project Badges: 
|
I am having the same issue. Apart from my CPU-only PCs, I have two PCs I am specifically trying to use with GPU (Windows, updated drivers) . One has NVidia GTX 440, the other has Intel UHD 520. Neither has received a single GPU job.
----------------------------------------In addition to setting custom profile options per WCG instruction, and additionally adding cc_config.xml to enable GPU, no effect. It shows init in the logs: (8/9/2021 12:12:22 PM | | OpenCL: Intel GPU 0: Intel(R) HD Graphics 520 From multiple searches through the support forum I see comments about GPU work being limited for OPN. Oddly though I find a post indicating that supposedly 80% of the project is GPU based. Therefore, how is it that minority CPU work is abundant and GPU is so sparse? I would expect for it to be inverted. CPU Available for download: 3,601 batches 2,950 batches in the last 30 days Average of 98.0 batches per day Estimated backlog: 36.6 days GPU Available for download: 6,338 batches In progress: 5,441 batches Completed: 60,876 batches 9,520 batches in the last 30 days Average of 317.0 per day I realize this is a loaded question, but hoping someone has an abridged answer. Thanks in advance. [Edit 1 times, last edit by william_az at Aug 9, 2021 8:12:37 PM] |
||
|
|
Falconet
Master Cruncher Portugal Joined: Mar 9, 2009 Post Count: 3297 Status: Offline Project Badges: 
|
william_az,
----------------------------------------1700 work units for GPUs are released every 30 minutes. That's low and they get sent very quickly to GPU devices. 80% of the work is done by GPUs but not 80% of tasks. A CPU task usually has 1 or 2 jobs inside which probably take around 1.5-3 hours in most computers, depending on how fast the CPU is. However, a GPU task can have 60, 80, 140 or more jobs inside while taking mere minutes on the fastest GPUs (3 minutes, for example). Because of that, 80% of the work is being done by GPUs but most tasks are still run on CPUs. I couldn't find any Nvidia card called GTX 440 but I found the Geforce GT 440. It seems that it only supports OpenCL 1.1 while the requirement for OPNG is OpenCL 1.2. So that won't work - it's a very slow card anyway. Regarding the Intel 520, OpenCL 1.2 seems to be supported - when BOINC requests WCG work, can you see it requesting Intel work on the Event Log? AMD Ryzen 5 1600AF 6C/12T 3.2 GHz - 85W AMD Ryzen 5 2500U 4C/8T 2.0 GHz - 28W AMD Ryzen 7 7730U 8C/16T 3.0 GHz [Edit 1 times, last edit by Falconet at Aug 9, 2021 8:30:05 PM] |
||
|
|
william_az
Cruncher Joined: Dec 10, 2008 Post Count: 3 Status: Offline Project Badges:
|
Falconet,
----------------------------------------Thank you very much for the reply. Much appreciated. I have read through quite a few GPU-related posts and "how-to's", and you explained that in a very clear way that made more sense. :-) Perhaps the issue is just that there is low GPU work. I just find it odd that my PC runs 24 hours a day, yet seems to miss every single GPU unit release On the PC with the Intel GPU, it does show that it is checking for work. Some of the log entries will show the CPU and GPU simultaneously checking for work, whereas some will show just the GPU requesting, but not receiving work. 8/9/2021 1:37:06 PM | World Community Grid | Sending scheduler request: To fetch work. 8/9/2021 1:37:06 PM | World Community Grid | Reporting 1 completed tasks 8/9/2021 1:37:06 PM | World Community Grid | Requesting new tasks for Intel GPU 8/9/2021 1:37:08 PM | World Community Grid | Scheduler request completed: got 0 new tasks Here is the line on client launch where it shows initialization: 8/9/2021 12:12:22 PM | | OpenCL: Intel GPU 0: Intel(R) HD Graphics 520 (driver version 27.20.100.8681, device version OpenCL 2.1 NEO, 3219MB, 3219MB available, 202 GFLOPS peak) In the Custom Profile, I have verified that all chipsets were set to yes. Just today I made a CC_Confnig.xml (which doesn't exist by default with Windows client install). It doesn't appear that the .xml config file is required - but I tried it anyway. No effect. (cc_config.xml file location: C:\ProgramData\Boinc ) This is what is in it: <cc_config> <log_flags> </log_flags> <options> <use_all_gpus>1</use_all_gpus> </options> </cc_config> Thanks again, and regards. [Edit 3 times, last edit by william_az at Aug 9, 2021 10:14:13 PM] |
||
|
|
Acibant
Advanced Cruncher USA Joined: Apr 15, 2020 Post Count: 126 Status: Offline Project Badges:
|
Have you seen the posts where people have set up clients to check for work every five minutes via scripting, then also use app_config.xml to run 8-10 simultaneously on a card? If someone has a high-end card they can complete all of those before the next forced update, though even then they will get nothing more often than not. While it's true there could be a possibility of misconfiguration on your part, it's far more likely that all the GPU jobs are getting grabbed up and consumed quickly. They ration the work units sent out, otherwise the researchers themselves wouldn't be able to keep up.
---------------------------------------- |
||
|
|
william_az
Cruncher Joined: Dec 10, 2008 Post Count: 3 Status: Offline Project Badges:
|
Acibant,
Thank you for the reply. Yes, I have seen a couple of posts about scripting on Linux machines, but not Windows. I might try again at setting up one of my GPU PCs with Linux. I made one attempt but had issues with the drivers when attempting to recompile the kernel. I believe you and Falconet are both correct about the GPU queue just being empty when my PCs check. Take care, william_az |
||
|
|
Tamagoch
Cruncher Ukraine Joined: Jul 21, 2007 Post Count: 33 Status: Offline Project Badges:
|
yes, indeed.... GPU task number is unable to fulfill a huge demand
as I'm using BOINC for testing everytime on different machines, it's quite rare I could get an OPNG workunit... at least I know the whole grid are doing really well to crunch everything |
||
|
|
Rizer579
Cruncher Joined: Jun 20, 2020 Post Count: 2 Status: Offline |
Thank you very much for your response. I've checked out what you said and I've not received any GPU work. BOINC does acknowledge in Event Log my GPU but seems no GPU tasks available every time my PC fetches. Can't quite understand why we'd want to crunch tasks with CPUs in 1.5 to 3 hours when we could do each one in minutes with GPUs.
|
||
|
|
Acibant
Advanced Cruncher USA Joined: Apr 15, 2020 Post Count: 126 Status: Offline Project Badges:
|
Can't quite understand why we'd want to crunch tasks with CPUs in 1.5 to 3 hours when we could do each one in minutes with GPUs. If I had my way all OPN work units would be GPU work units but they are trying to keep the people with Raspberry Pis that have ARM processors busy as there are no other current WCG projects for that CPU architecture running on Linux. They'd have the option to run Mapping Cancer Markers if they installed Android but that can be problematic in trying to run headless. |
||
|
|
|