Not receiving enough with WUs

Author	Message
Masse Send message Joined: 26 May 21 Posts: 4 Credit: 48,442,927 RAC: 37,501	Message 70873 - Posted: 15 Jun 2021, 14:10:55 UTC Got two nVidia K80 (but only one installed yet) about a week ago and have realized that if I shut down the network, I will run out of GPU WUs within 2-3 hours. Have tried to change the values, both in the account settings and in BOINC Manager, but ... It seems that the requested amount is related to the CPU, but not the GPU's. Is this problem related to MW@H or BOINC Manager.? Anyhow, it's sort of a 'nice' problem ... [ i9-14900K/128GB/A2000 + i7-11700K/128GB/A4000 + i5-10400T/32GB/UHD630 + i7-7700/32GB/A5000 + i7-3770K/32GB/P4 \| Ubuntu 22.04.4 LTS ] ID: 70873 · Rating: 0 · rate: / Reply Quote

Keith Myers Send message Joined: 24 Jan 11 Posts: 715 Credit: 555,629,588 RAC: 42,194	Message 70874 - Posted: 15 Jun 2021, 14:25:27 UTC - in response to Message 70873. Number of tasks requested depends on the BOINC settings for "days of work" and "additional days of work" with the maximum being 10/10. But the project limits the total amount of tasks carried by a host to ngpus X 300. So for your two cards you will be allotted 600 tasks maximum to be carried daily in your cache. ID: 70874 · Rating: 0 · rate: / Reply Quote

Masse Send message Joined: 26 May 21 Posts: 4 Credit: 48,442,927 RAC: 37,501	Message 70875 - Posted: 15 Jun 2021, 14:38:06 UTC - in response to Message 70874. Last modified: 15 Jun 2021, 14:59:45 UTC The K80(s) will then be waiting a lot, as I normally keep the network shut down at times when I'm not at home or asleep. Sad ... Edit: The K80 is a double GPU card, so it would be 900. The K80 will finish the majority of those in 'no time', as each GPU on the K80 finish every WU (de_modfit_8x_bundle4_4s_south4s_gapfix) in less then one minute. [ i9-14900K/128GB/A2000 + i7-11700K/128GB/A4000 + i5-10400T/32GB/UHD630 + i7-7700/32GB/A5000 + i7-3770K/32GB/P4 \| Ubuntu 22.04.4 LTS ] ID: 70875 · Rating: 0 · rate: / Reply Quote

Keith Myers Send message Joined: 24 Jan 11 Posts: 715 Credit: 555,629,588 RAC: 42,194	Message 70876 - Posted: 15 Jun 2021, 19:29:04 UTC - in response to Message 70875. Well then, your next choice would be to spoof the gpu count to increase the number of tasks allotted. Edit the coproc_info.xml file to clone the original card entry multiple times then lock the file afterwards so BOINC can not update or change your new values. ID: 70876 · Rating: 0 · rate: / Reply Quote

Vester Send message Joined: 30 Dec 14 Posts: 34 Credit: 909,998,366 RAC: 0	Message 70877 - Posted: 15 Jun 2021, 19:56:22 UTC I sent you a private message, Masse. ID: 70877 · Rating: 0 · rate: / Reply Quote

Masse Send message Joined: 26 May 21 Posts: 4 Credit: 48,442,927 RAC: 37,501	Message 70878 - Posted: 15 Jun 2021, 20:30:04 UTC - in response to Message 70876. Nice one! A 'quick-and-dirty', like that and are going to try it. Thank you! [ i9-14900K/128GB/A2000 + i7-11700K/128GB/A4000 + i5-10400T/32GB/UHD630 + i7-7700/32GB/A5000 + i7-3770K/32GB/P4 \| Ubuntu 22.04.4 LTS ] ID: 70878 · Rating: 0 · rate: / Reply Quote

Masse Send message Joined: 26 May 21 Posts: 4 Credit: 48,442,927 RAC: 37,501	Message 70879 - Posted: 16 Jun 2021, 16:50:02 UTC - in response to Message 70876. Have tested it. It only produce a number of GPU calculation errors. [ i9-14900K/128GB/A2000 + i7-11700K/128GB/A4000 + i5-10400T/32GB/UHD630 + i7-7700/32GB/A5000 + i7-3770K/32GB/P4 \| Ubuntu 22.04.4 LTS ] ID: 70879 · Rating: 0 · rate: / Reply Quote

mmonnin Send message Joined: 2 Oct 16 Posts: 167 Credit: 1,008,062,758 RAC: 1,668	Message 70900 - Posted: 23 Jun 2021, 2:59:14 UTC I always run several BOINC clients when running MW to allow for 300 tasks on each client. Radeon VII spits them out in a couple of hours. Stagger the clients a bit and the card will always stay busy. ID: 70900 · Rating: 0 · rate: / Reply Quote