A "saw" pattern of the GPU load. Can someone help indentify the cause?
28 Sep 2012 12:56:28 UTC
Topic 196543
(moderation:
Hello! Have a strange "saw" pattern of GPU activity when boinc is active. See a picture here, please: http://clck.ru/2MCDx
Can someone help to indentify the cause? The scale of the graph is 10 seconds (so each increment in time scale is 10 seconds). So every 1-2minutes we have a short GPU drop of activity.
What can cause such a problem?
A "saw" pattern of the GPU load. Can someone help indentify the
)
PS can someone tell me how to insert a picture into the body of the message, not a link?
Each BRP4 task - if that's
)
Each BRP4 task - if that's the type you're running - is actually a 'package' of eight tasks. That's why you download 8 data files, and upload 8 result files, for each job.
I see the GPU pausing briefly, seven times per task, at the changeover between between the sub-tasks.
It is possible to embed individual images (though not whole websites like that one) on these boards with [img] tags - but please don't do it with that image, it's far too large to be accommodated here.
M-m-m... U mean this pattern
)
M-m-m... U mean this pattern is normal? I don't remember such a strange pattern on my previous R6850... Changed VC recently to HD7850
RE: M-m-m... U mean this
)
I've seen that kind of pattern on GPU usage when upgraded my GPUs for faster ones without reserving enough CPU cores to feed them...
Strange... I have i7-2600K @
)
Strange... I have i7-2600K @ 4400. 4 cores, 8 virtual
Need to reserve 1 core for GPU?
RE: Strange... I have
)
Reserving one core ussually helps a lot, in most cases reserving more than one is even better, but thats something you will need to test on your own system as it depends on a lot of other things like OS, other apps, what other things you do in the hosts, and even it can change if you live in rainy area... ;D
If you do not reserve a core, then all the cores will be always bussy and when the GPU needs CPU help (and in Einstein GPUs requiere a considerable help from the CPU), the CPU cores will be "overloaded" and the OS will need to do a lot of switching between the running apps/threads to run them all and meanwhile the GPUs will be waiting...