There are some findings on SETI forums that affinity can improve GPU load if 
only 1 CPU bound with GPU app (under full CPU load w/o CPU core reservation).
But very possible that binding all GPU apps (if there are few GPU devices or 
few instances per device or both) to the same CPU would be not so efficient as 
to bind each instance to each own CPU. But for such load balancing app should 
know how many another instances are in fly already. Considering GPU apps from 
different projects it would be better to keep track of GPU app instances in fly 
with their GPU assignments on BOINC level rather than on app level. Hence my 
question.



Понедельник, 18 марта 2013, 12:50 UTC от Richard Haselgrove 
<[email protected]>:
>Under what circumstances would an app need to know?
>
>I don't think there is a way at the moment: if you're putting in a feature 
>request - eg for a converse of the MT --nthreads x cmdline construct - I 
>suggest you put in some more explanation.
>
>>----------------------------------------------------------------------
>>From: Raistmer the Sorcerer <[email protected]>
>>To: [email protected] 
>>Sent: Monday, 18 March 2013, 12:19
>>Subject: [boinc_dev] How to find number of instances of app?
>>
>>BOINC provides GPU device number for app via --device switch.
>>But how it informs app how many app instances it will run on single GPU 
>>device ? 
>>Is there some way for app to know this w/o implementing own counting scheme ?
>> 
>>_______________________________________________
>>boinc_dev mailing list
>>[email protected]
>>http://lists.ssl.berkeley.edu/mailman/listinfo/boinc_dev
>>To unsubscribe, visit the above URL and
>>(near bottom of page) enter your email address.
>>

_______________________________________________
boinc_dev mailing list
[email protected]
http://lists.ssl.berkeley.edu/mailman/listinfo/boinc_dev
To unsubscribe, visit the above URL and
(near bottom of page) enter your email address.

Reply via email to