Crunching Cuda = Killing CPU jobs?

Questions and Answers : GPU applications : Crunching Cuda = Killing CPU jobs?
Message board moderation

To post messages, you must log in.

AuthorMessage
VSeven

Send message
Joined: 25 Nov 99
Posts: 9
Credit: 4,929,679
RAC: 0
United States
Message 886522 - Posted: 20 Apr 2009, 0:47:35 UTC

Intel Core i7 920 @ 3.2Ghz. HT turned off. 12Gb RAM. GTX260 55nm w 896Mb of ram.

I "was" processing Seti packets at about a 1h 54m clip, consistantly for weeks now. I upgraded to 6.6.20 and got a hundred or so Cuda WU's downloaded. The Cuda WU's are flying along at like 10 - 14 minutes per WU but now the CPU WU's are taking around 10 hours each.

Is this normal to have that much cpu "overhead" for the Cuda apps?
ID: 886522 · Report as offensive
Andy Williams
Volunteer tester
Avatar

Send message
Joined: 11 May 01
Posts: 187
Credit: 112,464,820
RAC: 0
United States
Message 886578 - Posted: 20 Apr 2009, 6:18:21 UTC - in response to Message 886522.  

It sounds like you're running Astropulse work units on the CPU. Is that the case? If so, let 'em go. The credits per flop are better for Astropulse.
--
Classic 82353 WU / 400979 h
ID: 886578 · Report as offensive
Profile skildude
Avatar

Send message
Joined: 4 Oct 00
Posts: 9541
Credit: 50,759,529
RAC: 60
Yemen
Message 886665 - Posted: 20 Apr 2009, 14:31:04 UTC
Last modified: 20 Apr 2009, 14:40:15 UTC

I see that you aren't running rasitmers opt app v10. The VLAR deletion rule on the app may be unnecessary for 6.6.20 but it also optimizes the CPU which you aren't getting with the standard app.


In a rich man's house there is no place to spit but his face.
Diogenes Of Sinope
ID: 886665 · Report as offensive
Profile perryjay
Volunteer tester
Avatar

Send message
Joined: 20 Aug 02
Posts: 3377
Credit: 20,676,751
RAC: 0
United States
Message 886681 - Posted: 20 Apr 2009, 15:00:31 UTC - in response to Message 886665.  

Skildude, 6.6.20 still gives VLARs as 6.08s so the killer is still very much needed. It looks like Raistmer is working on something to rebrand them so they will be sent to the CPU as 6.03s but it isn't quite there yet.


PROUD MEMBER OF Team Starfire World BOINC
ID: 886681 · Report as offensive
VSeven

Send message
Joined: 25 Nov 99
Posts: 9
Credit: 4,929,679
RAC: 0
United States
Message 886801 - Posted: 20 Apr 2009, 22:18:30 UTC

Actually I had a bunch of regular WU's queued up when I upgraded (what should have been about a days worth but took 4 or 5 because of the cuda app) so those had to crunch through. As of right now I'm crunching 4 AP WU's and I have queued up what looks to be about 230 Cuda WU's.

No I'm not running a "optimized" app. I update BOINC every so often and let it do it's thing, don't really have time to hunt down optimized apps and keep them upgraded.
ID: 886801 · Report as offensive
Profile zoom3+1=4
Volunteer tester
Avatar

Send message
Joined: 30 Nov 03
Posts: 65746
Credit: 55,293,173
RAC: 49
United States
Message 887618 - Posted: 23 Apr 2009, 15:48:01 UTC - in response to Message 886681.  
Last modified: 23 Apr 2009, 15:50:50 UTC

Skildude, 6.6.20 still gives VLARs as 6.08s so the killer is still very much needed. It looks like Raistmer is working on something to rebrand them so they will be sent to the CPU as 6.03s but it isn't quite there yet.

So far You must have 500 or less WU's in Your cache or the Perl script won't do the job. Yesterday I had about 877, Now I have 668. Of course to do this I had to set Boinc to NNT and I've since run out of 6.08 WU's, I just have 6.03 WU's now or about 6.5 days worth.
The T1 Trust, PRR T1 Class 4-4-4-4 #5550, 1 of America's First HST's
ID: 887618 · Report as offensive

Questions and Answers : GPU applications : Crunching Cuda = Killing CPU jobs?


 
©2024 University of California
 
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.