Message boards :
Number crunching :
Optimizing for happiness with NVIDIA GPU
Message board moderation
Author | Message |
---|---|
ralphw Send message Joined: 7 May 99 Posts: 78 Credit: 18,032,718 RAC: 38 |
The story so far:
|
TBar Send message Joined: 22 May 99 Posts: 5204 Credit: 840,779,836 RAC: 2,768 |
Well, the 7.08 (opencl_nvidia_sah) is about as new as you can get, it was released less than 2 months ago. The AP App on Main is also the newest. To try the other Apps you will have to go to Anonymous platform and on Linux it's all manual. You will have to round up the Apps and make your own app_info.xml I would suggest the Linux CUDA App, in my experience it's a little slower on the shorties but makes up for it on the longer tasks which rarely take longer than 900 seconds on a 750Ti. Here is a 750Ti running the CUDA App, http://setiathome.berkeley.edu/results.php?hostid=7772630&state=4&appid=11 You can find it here, http://www.arkayn.us/forum/index.php?action=tpmod;dl=item132 I can't seem to find the AP App, so, you'll have to use the one on Main. The CPU Apps are better than on Main, they are here; http://lunatics.kwsn.info/index.php?action=downloads;cat=1 That should get you started... |
Grant (SSSF) Send message Joined: 19 Aug 99 Posts: 13727 Credit: 208,696,464 RAC: 304 |
I think I can achieve 10,000 WUs/day As poorly implemented as Credit New is, it's a better indicator of work done than the number of WUs. The number of WUs processed per hour is useful when optimising the number of WUs to crunch at a time (generally 2, maybe 3 for highend cards), but you have to be careful you're comparing the same things- shorties v VLARs just can't be compared. You have to compare shortie v shortie & VLAR v VLAR runtimes. Then you have to determine if the increased shortie crunching time is offset (enough) by the faster VLAR crunching times when running 2 or 3 at a time. For my GTX 750Ti's 2 at a time is best. 3 at a time is good for VLARs, but shorter run time WUs the processing time increases significantly & more than offsets any improved VLAR crunching times. So 2 at a time is best for me. And AP WUs take much, much longer to crunch than MB WUs, but you get much, much higher levels of credit than MB for crunching them. Hence, as messed up as it is, RAC is the best indicator. Unfortunately it's a lagging indicator- with major hardware/software changes it takes 6-8 weeks to settle down to it's new "normal" level. And that's if there are no Seti system outages. Grant Darwin NT |
Richard Haselgrove Send message Joined: 4 Jul 99 Posts: 14650 Credit: 200,643,578 RAC: 874 |
I think you're being a bit literal there, Grant. 10,000 WUs per day would be one every 8.6 seconds - not even Perano's top host manages that. If you compare the OP's "processing about 6800 workunits per day" with his current RAC, we may just have a simple 'lost in translation' between credits and WUs. |
Grant (SSSF) Send message Joined: 19 Aug 99 Posts: 13727 Credit: 208,696,464 RAC: 304 |
10,000 WUs per day would be one every 8.6 seconds - not even Perano's top host manages that. Ah, I've never even considered how many WUs per hour per card etc would be done in a day. Something to aim for with the application optimisations. Grant Darwin NT |
ralphw Send message Joined: 7 May 99 Posts: 78 Credit: 18,032,718 RAC: 38 |
Yes, I meant Credit - Recent Average Credit (RAC). Not WorkUnits. 10,000 Workunits per day would mean I wouldn't need my furnace to heat my home anymore! |
©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.