Message boards : Number crunching : Not Happy
Author | Message |
---|---|
Do this because it is worthwhile; however, I am also competitive and results driven. I know the tail should not wag the dog but,I bought a GTX 580 solely for this project. | |
ID: 26713 | Rating: 0 | rate: / Reply Quote | |
Yes, your 450 is a bit on the slow side but you already new that :-) | |
ID: 26714 | Rating: 0 | rate: / Reply Quote | |
I had that problem before, reduce your cache in BOINC to 0.01 or 0.05 days and it will only request a new task right before the current one finishes. | |
ID: 26721 | Rating: 0 | rate: / Reply Quote | |
Thanks for the hints. I have adjusted my caches, good tip. However, the problem is not with my rig. I regularly turn in over 300,000 points per day. I think I am #1 in RAC for KWSN team by a long margin. What's up with that? . That is why I am unhappy. My EVGA 660 Ti will be here in 2 days, watch this space | |
ID: 26739 | Rating: 0 | rate: / Reply Quote | |
even if you free one cpu core for gpugrid? | |
ID: 26742 | Rating: 0 | rate: / Reply Quote | |
My EVGA 660 Ti will be here in 2 days, watch this space SC or FTW? Keep me posted, as I am wondering if I shall ask a friend of mine to bring me one from the USA in October? | |
ID: 26745 | Rating: 0 | rate: / Reply Quote | |
gdf, I have 2 of 4 cores free. | |
ID: 26746 | Rating: 0 | rate: / Reply Quote | |
even if you free one cpu core for gpugrid? No. These PAOLA_3EKO_8LIGANDS won't use a full CPU core. They can easily spot from my host's tasklist, as their CPU time is much less than the corresponding GPU time. | |
ID: 26748 | Rating: 0 | rate: / Reply Quote | |
afaik SWAN_SYNC is no longer necessary for max performance. Somebody know for sure? | |
ID: 26749 | Rating: 0 | rate: / Reply Quote | |
660 Ti installed and running with driver 306.02. At present a PAOLA_3EKO_8LIGANDS has completed approx 16.5% in 3 hrs So we can expect the WU to finish in a little over 19 hours. | |
ID: 26752 | Rating: 0 | rate: / Reply Quote | |
afaik SWAN_SYNC is no longer necessary for max performance. Somebody know for sure? This variable is handled internally by the CUDA4.2 client. For example, the CUDA4.2 client runs with low CPU usage on my Fermi GPUs without any noticeable performance loss, but it uses a full CPU core on Kepler based cards, except for the PAOLA_3EKO_8LIGANDS series of workunits. I think some setting (regarding the SWAN_SYNC behavior) was wrong upon generating these workunits. For example: while the NATHAN_RPS1120528 and NATHAN_RPS1120801_N use very little CPU time on Fermi GPUs, and use the same amount of CPU and GPU time on Kepler GPUs, these PAOLA_3EKO_8LIGANDS use much more CPU time on Fermi GPUs, roughly the same amount of CPU time as on Kepler GPUs (but it's much less than the GPU time taken). | |
ID: 26753 | Rating: 0 | rate: / Reply Quote | |
660 Ti installed and running with driver 306.02. At present a PAOLA_3EKO_8LIGANDS has completed approx 16.5% in 3 hrs So we can expect the WU to finish in a little over 19 hours. Supporting my theory explained in my previous post, the lesser Kepler GPUs suffer less performance loss with these PAOLA_3EKO_8LIGANDS than the faster ones. | |
ID: 26754 | Rating: 0 | rate: / Reply Quote | |
Lesser? Careful you might offend my brand new GPU! | |
ID: 26755 | Rating: 0 | rate: / Reply Quote | |
Sorry for being dubious, I didn't mean to offend your brand new GPU. Your GTX660Ti's progress report called my GT 630's performance in my mind. (it's not a Kepler, but let's consider it a typo) :) | |
ID: 26766 | Rating: 0 | rate: / Reply Quote | |
Message boards : Number crunching : Not Happy