Message boards :
Number crunching :
Can't get work for NVIDIA GPU
Message board moderation
Author | Message |
---|---|
Send message Joined: 5 Mar 09 Posts: 6 Credit: 60,945,378 RAC: 0 |
Hi, I haven't crunched Milkway for a few years and just getting back to it. I have 3 machines set to do GPU work only: 1. 766835 AMD 470 + NVIDIA GTX 1060 2. 766724 AMD 460 + NVIDIA GTX 750ti 3. 766830 2 x NVIDIA GTX 1060 All have cc_config.xml in BOINC directory specifying use_all_gpus= 1 Initially I could not get any GPU work at all: 26/02/2018 10:46:52 AM | Milkyway@Home | Sending scheduler request: To fetch work. 26/02/2018 10:46:52 AM | Milkyway@Home | Requesting new tasks for NVIDIA GPU and AMD/ATI GPU 26/02/2018 10:46:54 AM | Milkyway@Home | Scheduler request completed: got 0 new tasks however AMS WUs are now flowing fine. On Computer 1, 1 NVIDIA WU 1570793768 was fetched and successfully processed on Feb 26th. On Computer 2, no NVIDIA tasks have been fetched or processed. On Computer 3, 3 NVIDIA tasks were fetched and successfully processed on Feb 25th (3 days ago) but nothing since The detailed records of the WUs don't seem to be available any more but you can see three tasks have been completed: Application details for host 766830 Computer 3 is continuing to try to fetch work: 28/02/2018 4:51:27 AM | Milkyway@Home | Sending scheduler request: To fetch work. 28/02/2018 4:51:27 AM | Milkyway@Home | Requesting new tasks for NVIDIA GPU 28/02/2018 4:51:30 AM | Milkyway@Home | Scheduler request completed: got 0 new tasks I've run out of ideas :-( Thanks, D |
Send message Joined: 23 Aug 11 Posts: 7 Credit: 1,217,109,541 RAC: 0 |
Yep same here, so far I tried 3 Nvidia based PC's and none of them can get work either. This one with a pair of 970"s http://milkyway.cs.rpi.edu/milkyway/show_host_detail.php?hostid=634688 This one with a single 980 ti. http://milkyway.cs.rpi.edu/milkyway/show_host_detail.php?hostid=634696 Another with a pair of 1080ti's http://milkyway.cs.rpi.edu/milkyway/show_host_detail.php?hostid=551198 Tried resetting & re-attaching but it made no difference. It's a little early to tell but so far this seems to only affect newly attached PC's & PC's that haven't crunched Milky way for a while. I last crunched here a month or 2 ago. Any help? |
Send message Joined: 23 Aug 11 Posts: 7 Credit: 1,217,109,541 RAC: 0 |
Still nothing and not one word one the problem from the admin? |
Send message Joined: 5 Mar 09 Posts: 6 Credit: 60,945,378 RAC: 0 |
One of 3 PCs has now started getting work for NVIDA GPUs. This one also has an AMD GPU. The other two still sitting idle as far as NVIDIA is concerned (one also has an AMD GPU which is getting work without problems. The other has 2 NVIDIA GPUs and is sitting idle. It got 1 WU yesterday, which was successfully processed and validated). |
Send message Joined: 30 Sep 07 Posts: 17 Credit: 269,654,203 RAC: 0 |
I see a common problem here with those who can not get work. In my case, I have not worked MilkyWay for over a year. I now have 16 computers with a variation of the latest AMD and NVIDIA GPU's and I can not get work on any of them. I checked my settibgs, Detached, reattached, installed the latest driver, nothing helped. The machines run most other GPU projects, when work available. |
Send message Joined: 1 Sep 09 Posts: 2 Credit: 88,165,232 RAC: 0 |
I have one pc with 2 Nvidia GTX 470's running on an amd 1090T processor windows 7, that has no issues getting and processing work. On the other hand I have one intel i7-4790K processor with an Nvidia 980ti windows 10 not getting any work except cpu tasks. Also have an amd 8150 windows 10 and it started getting work earlier today. Still searching for a resolution to my one pc not getting gpu work. Note I have tried the usual fixes listed above in other posts. |
Send message Joined: 1 Sep 09 Posts: 2 Credit: 88,165,232 RAC: 0 |
Now all Nvidia cards seem to be getting work. Hmmm... |
Send message Joined: 8 May 09 Posts: 3319 Credit: 520,285,017 RAC: 19,707 |
Now all Nvidia cards seem to be getting work. Hmmm... They made a change that was also preventing people with bad pc's from getting workunits, unfortunately that setting was also preventing new pc's from getting work too, they changed that setting back and everyone should be getting workunits now. |
Send message Joined: 29 Jul 14 Posts: 19 Credit: 3,451,802,406 RAC: 8 |
I will mention that the developers are reducing work units to GPUs without very high double precision capability. Pretty much all of the NVIDIA GPUs mentioned in this thread have *very low* FP64 (double precision) capability, which would explain why you haven't been getting many work units. You should allocate your GPUs elsewhere. |
Send message Joined: 2 Oct 16 Posts: 167 Credit: 1,005,828,164 RAC: 54,613 |
I will mention that the developers are reducing work units to GPUs without very high double precision capability. Pretty much all of the NVIDIA GPUs mentioned in this thread have *very low* FP64 (double precision) capability, which would explain why you haven't been getting many work units. You should allocate your GPUs elsewhere. Only GPUs without ANY FP64 capability and thus produce 100% errors. If it can complete work there's no reason to ban hosts. Those that said they could not get work before have received work since. |
Send message Joined: 13 Sep 16 Posts: 14 Credit: 95,008,484 RAC: 0 |
Hi all, since one day my TITANz produce only "calculation error". There is no SLI only double precision is turn on in the settings of the Nvidia control panel. At the moment i run WIN 7 Enterprise 64 Bit with following driver "382.33-desktop-win8-win7-64bit-international". Any newer driver is not working. I use a Nvidia 1070, TITAN X and a TITAN Z in one computer. If i use only the TITAN Z, everything works well. Has anyone experiences with mixed GPU system? |
Send message Joined: 8 May 09 Posts: 3319 Credit: 520,285,017 RAC: 19,707 |
Hi all, SLI does NOT work in any Boinc Project to my knowledge and is not even on the radar right now. That means you have to treat each gpu as a standalone one, so a line in your cc_config.xml file to <use_all_gpus> is needed like this: <cc_config> <options> <use_all_gpus>1</use_all_gpus> </options> </cc_config> After that you can add more lines between the <options> </options> lines to exclude one or more of your other gpus for any project you chose by changing the url line, for example this setup excludes gpu 1 from Milikyway: <exclude_gpu> <url>http://milkyway.cs.rpi.edu/milkyway/</url> <device_num>1</device_num> </exclude_gpu> Adding this line would then exclude gpu 1 from Collatz: <exclude_gpu> <url>http://boinc.thesonntags.com/collatz/</url> <device_num>1</device_num> </exclude_gpu> Thus letting it run a 3rd project of your choosing. If you do this though you will need exclude both gpu 0 and gpu 2 from their respective projects. Since you have 3 gpu's they will be numbered 0, 1 and 2, when you first startup Boinc go to the Event Log and it will tell you which gpu is which. If you do want all 3 to crunch here at MW you may have to reload the drivers software for each different gpu, sometimes you do and sometimes you don't. A Dummy Plug can also be required for any gpu not connected to a monitor on boot-up in Windows because Windows will try and save resources on bootup and if no monitor is connected it can turn that gpu off. A way to make your own is here: https://www.techpowerup.com/forums/threads/30-second-dummy-plug.87670/ I have had to use these in some machines but not others, it's hit or miss for me. If you plug in a powered on monitor to each gpu on bootup these are not needed as only during bootup will Windows turn them off. |
Send message Joined: 13 Sep 16 Posts: 14 Credit: 95,008,484 RAC: 0 |
look here https://milkyway.cs.rpi.edu/milkyway/forum_thread.php?id=4632 Ich bin hier weil es BOINC gibt |
©2024 Astroinformatics Group