Resource share for max efficiency

Message boards : Graphics cards (GPUs) : Resource share for max efficiency

Author	Message
JAMC Send message Joined: 16 Nov 08 Posts: 28 Credit: 12,688,454 RAC: 0 Level Scientific publications	Message 3874 - Posted: 18 Nov 2008 \| 13:35:28 UTC
	Hello: I have just attached to this project and completed (2) WU's and have some questions... System- Q6600 OC'd to 3.15, XP Home, Evga 8800GT (512mb), (2) gb RAM I have GPUgrid assigned 23% resource share (BOINC Mgr.> Projects> Resource share) and the other two projects equal 77%. This is a dedicated 24/7 cruncher( some email and web browsing) and I wish to get the maximum computing done on all 4 cores. The first two WU's completed in 26,000 - 27,000s but I noticed that the WU run time was longer than elapsed real time ie., WU started at 12:00pm did not finish at 7:30pm as I expected but took longer. Is this due to system demands or should I expect them to match and get (3) WU's done in 24 hour period? Will increasing the resource share have a positive effect or no effect on GPUgrid WU's... will such a change negatively impact the other projects or do they automatically pick up any CPU resources not used by GPUgrid? Thanks!
	ID: 3874 \| Rating: 0 \| rate: / Reply Quote

Krunchin-Keith [USA] Send message Joined: 17 May 07 Posts: 512 Credit: 111,288,061 RAC: 0 Level Scientific publications	Message 3875 - Posted: 18 Nov 2008 \| 14:22:10 UTC - in response to Message 3874.
	Hello: I have just attached to this project and completed (2) WU's and have some questions... System- Q6600 OC'd to 3.15, XP Home, Evga 8800GT (512mb), (2) gb RAM I have GPUgrid assigned 23% resource share (BOINC Mgr.> Projects> Resource share) and the other two projects equal 77%. This is a dedicated 24/7 cruncher( some email and web browsing) and I wish to get the maximum computing done on all 4 cores. The first two WU's completed in 26,000 - 27,000s but I noticed that the WU run time was longer than elapsed real time ie., WU started at 12:00pm did not finish at 7:30pm as I expected but took longer. Is this due to system demands or should I expect them to match and get (3) WU's done in 24 hour period? Will increasing the resource share have a positive effect or no effect on GPUgrid WU's... will such a change negatively impact the other projects or do they automatically pick up any CPU resources not used by GPUgrid? Thanks! The time you refer to is CPU used time, not wall clock time. It does not include time while cpu is busy doing somethign else. The cpu is shared with another task and the percentage is not 100% for the cpu/gpu task, thus the difference. See http://www.gpugrid.net/forum_thread.php?id=500 where I outlined this performance. You also need to see the gpu elapsed time, to see this you need to drill down into your results. This will probably be closer to your wall clock time, but not exact. Other applciations can slow down the cpu, thus the gpu has to wait for the cpu to become free and poll it. Resource share is for cpu projects. It has some effect on cpu/gpu projects, but I'm not sure exactly what. Currently the cpu/cpu task is suppose to be treated as a high priority task, and run regardless of resource share, meaning that a gpu task should always be running. It may only matter if you had two gpu projects, eg you could run one 25% and one 75%. In the future there will separate prefs for cpu and cpu/gpu projects which may make things easier to schedule and control and understand. At this time things are so new a lot of the prefs are shared with cpu projects and some are not used.
	ID: 3875 \| Rating: 0 \| rate: / Reply Quote

Message boards : Graphics cards (GPUs) : Resource share for max efficiency

//