Advanced search

Message boards : Graphics cards (GPUs) : Performance: 4 CPUs -vs- 4+1 (9800gtx+)

Author Message
Profile JStateson
Avatar
Send message
Joined: 31 Oct 08
Posts: 186
Credit: 3,387,420,562
RAC: 1,239,778
Level
Arg
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 5181 - Posted: 3 Jan 2009 | 2:51:49 UTC

Ran a quick test that indicated I am better off dedicating a full CPU to the GPU. This is how I figured it.

I observed there were 15,000 seconds of CPU time available in 4+1 mode. That corresponds to an average of 800 GPUGRID credits (the 15k secs). I could not find any project that would give me 800 credits in 15,000 seconds.

MarkJ
Volunteer moderator
Volunteer tester
Send message
Joined: 24 Dec 08
Posts: 738
Credit: 200,909,904
RAC: 0
Level
Leu
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 5189 - Posted: 3 Jan 2009 | 11:39:37 UTC - in response to Message 5181.
Last modified: 3 Jan 2009 | 11:43:47 UTC

Ran a quick test that indicated I am better off dedicating a full CPU to the GPU. This is how I figured it.

I observed there were 15,000 seconds of CPU time available in 4+1 mode. That corresponds to an average of 800 GPUGRID credits (the 15k secs). I could not find any project that would give me 800 credits in 15,000 seconds.



I can't see why you'd need to dedicate a core to it. At the moment i'm running 4+1 using BOINC 6.5.0 without needing a cc_config.xml file. It runs the feeder task at "below normal" priority. I find that if I increase its priority to "normal" the run times come down (about 10 hours on a 9800GT). Obviously there is an impact on the cpu tasks running, but only a minimal effect, so you still get other work units processed on the cpu without having to dedicate a core. When the 6.56 app gets corrected the impact should be even less.

ExtraTerrestrial Apes
Volunteer moderator
Volunteer tester
Avatar
Send message
Joined: 17 Aug 08
Posts: 2705
Credit: 1,311,122,549
RAC: 0
Level
Met
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 5215 - Posted: 3 Jan 2009 | 19:26:52 UTC

Interesting, in your case the GPU really suffers in 4+1 compared to 3+1 (labeled 4 in your image). In the case of such a huge difference you're really better off with 3+1.

In my case the 9800GTX+ goes from 49 ms/step to 55 - 60 ms/step when switching from 3+1 to 4+1 and even with this smaller difference I figured running 3+1 would be more effective. And the faster the card the more pronounced the difference becomes.

MrS
____________
Scanning for our furry friends since Jan 2002

Post to thread

Message boards : Graphics cards (GPUs) : Performance: 4 CPUs -vs- 4+1 (9800gtx+)

//