Questions and Answers :
Wish list :
CPU
Message board moderation
| Author | Message |
|---|---|
|
Send message Joined: 7 Aug 22 Posts: 24 Credit: 20,845,775 RAC: 5,230 |
Why don't MT tasks use all available CPU cores and threads? Only 16 CPUs? In the Max # of threads for each MilkyWay@home task settings, you can set up to 256 cores and threads for ST tasks, which seems to be normal, but MT is limited to only 16 cores and threads... Or am I missing something and is it possible to make, for example, 2 tasks together and 10 cores and threads each so that the tasks are MT and not ST? |
|
Send message Joined: 19 Jul 10 Posts: 812 Credit: 20,951,030 RAC: 6,123 |
Or am I missing something and is it possible to make, for example, 2 tasks together and 10 cores and threads each so that the tasks are MT and not ST?Yes, if you set "Max # of threads for each MilkyWay@home task" to 10 (and leave "Max # of simultaneous MilkyWay@home tasks" preferably at "no limit"), than you can run 2 10-thread tasks on your CPU with 20 threads. Unfortunately the server isn't smart enough to set automatically a sensible amount of threads per task, the default works only for CPUs with 16 or less threads (or integer multiple of 16 like 32, 48, 64 etc.).
|
|
Send message Joined: 7 Aug 22 Posts: 24 Credit: 20,845,775 RAC: 5,230 |
It doesn't work. I set 2.10 in the settings, but it still uses 16 cores and threads, and only calculates one task. |
|
Send message Joined: 7 Aug 22 Posts: 24 Credit: 20,845,775 RAC: 5,230 |
The project had to be restarted after each change in the number of cores and threads. Thanks for the help. |
|
Send message Joined: 7 Aug 22 Posts: 24 Credit: 20,845,775 RAC: 5,230 |
But the maximum number of cores and threads per task still doesn't go above 16, which is bad, of course. |
|
Send message Joined: 19 Jul 10 Posts: 812 Credit: 20,951,030 RAC: 6,123 |
I've read on PrimeGrid, that there's a bug in BOINC 8.2.x, which causes that, people were complaining about similar issues there too. Downgrade to 7.24.1 and everything will work as it should, BOINC 8 is just crap with annoying Docker messages confusing everyone and other issues while it doesn't have any useful features, at least I could not find any for the projects I run and Milkyway is one of them.
|
|
Send message Joined: 7 Aug 22 Posts: 24 Credit: 20,845,775 RAC: 5,230 |
I reinstalled boinc and cleared any remaining system files, but it still didn't help. I tried setting it to "no restrictions" and it still selects 16. I also tried forcing it to 20, but it still selects 16. |
|
Send message Joined: 7 Aug 22 Posts: 24 Credit: 20,845,775 RAC: 5,230 |
I should have upgraded my CPU to an i9 14900k, it would have been able to handle two tasks on 16 cores/threads :) |
|
Send message Joined: 19 Jul 10 Posts: 812 Credit: 20,951,030 RAC: 6,123 |
I tried setting it to "no restrictions" and it still selects 16.This is expected, default is 16. I also tried forcing it to 20, but it still selects 16.On new tasks received after the change? Tasks received before the change will show in BOINC Manager what they were supposed to run on when you got them, they will still use the new value (except the already running ones unless you restart the BOINC client). This task was running on 10 threads, so I'd say it worked, it was just not displayed in BM. <core_client_version>8.2.8</core_client_version> <![CDATA[ <stderr_txt> <search_application> milkyway_nbody 1.94 Windows x86_64 double OpenMP, Crlibm </search_application> Using OpenMP 10 max threads on a system with 20 processors Running MilkyWay@home Nbody v1.94 Optimal Softening Length = 0.002068807820259 kpc, Upper bound = 0.017975612260566 kpc, Lower bound = 0.000238098471147 kpc Dwarf Initial Position: [-18.519759334690484,41.773119675543214,2.521461353061421] Dwarf Initial Velocity: [28.335193244637974,167.381900791132637,-93.432151455203140] Initial LMC position: [79.091156289642782,454.189801093785604,-196.850190073377576] Initial LMC velocity: [-24.277471862838940,-168.921658317774501,38.174726064010969] Initial body data written to 'initial.out' <search_likelihood>-102.112417526875760</search_likelihood> <search_likelihood_EMD>-16.136140380795954</search_likelihood_EMD> <search_likelihood_Mass>-2.146054680655134</search_likelihood_Mass> <search_likelihood_Beta>-21.231419499514832</search_likelihood_Beta> <search_likelihood_BetaAvg>-22.865070427951220</search_likelihood_BetaAvg> <search_likelihood_VelAvg>-22.501601939967848</search_likelihood_VelAvg> <search_likelihood_Dist>-13.939768346258273</search_likelihood_Dist> <search_likelihood_Momentum>-3.292362251732496</search_likelihood_Momentum> strftime() failed called boinc_finish(0) </stderr_txt> ]]> So set it back to 10, allow new work, and if you allow 100% of CPUs, two MW tasks should start running. Since you have no in progress tasks, it should also be displayed properly in BOINC Manager.
|
©2026 Astroinformatics Group