Message boards :
Number crunching :
Getting most from S0G: Fine tuning r3500
Message board moderation
Author | Message |
---|---|
Stephen "Heretic" Send message Joined: 20 Sep 12 Posts: 5557 Credit: 192,787,363 RAC: 628 |
. . Hi everyone, . . I feel S0G has a lot to offer and I want to get the most from it that I can. It would be great to get runtimes as low as possible and I am trying to find the best settings to achieve that. . . I would love to hear which settings have been tried by others and with what results. Stephen . |
Stephen "Heretic" Send message Joined: 20 Sep 12 Posts: 5557 Credit: 192,787,363 RAC: 628 |
Rig:. . . Core 2 Duo 3GHz E7600 with NVidia GT730 (2GB GDDR5 Ram) OS: . . . Windows 10 Pro 64 bit App: . .. SoG r3500 Command:. -high_prec_timer -use_sleep -sbs 384 -tt 90 -hp . . These settings are giving 98% GPU utilisation making running multiple tasks pointless. When I first tried r3500 NormAR Arecibo WUs took about 39 mins, approx 5 mins longer than CUDA50 tasks running singly but I had been running doubles with CUDA50 for average runtimes of 30 mins, so 9 mins longer for me. Guppis under CUDA50 were the fly in the ointment taking about 2 hours running doubles (2.5 for BLC5....) but under r3500 running singles they now take about 49 mins (about 62 mins for BLC5) Which is a big improvement. And the runtimes are very consistent. . . Those times are for the original settings (-sbs 256 -tt 60 and no -hp) but since adopting the new settings I have either hit the longest continuous run of halflings (shorty WUs) that I have ever encountered or normAR Arecibo tasks are now only taking 18 mins. I find that somewhat incredible so I am still waiting for the "run" to end and go back to the original times. But these WUs are from many different series (ie very different names) which makes it seem remote that they might all be halflings. Stephen . |
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
+ remaining tuning line from ReadMe (for oclFFT tuning for example) SETI apps news We're not gonna fight them. We're gonna transcend them. |
Stephen "Heretic" Send message Joined: 20 Sep 12 Posts: 5557 Credit: 192,787,363 RAC: 628 |
+ remaining tuning line from ReadMe (for oclFFT tuning for example) . . OK I will add that to the command line. Stephen |
Stephen "Heretic" Send message Joined: 20 Sep 12 Posts: 5557 Credit: 192,787,363 RAC: 628 |
+ remaining tuning line from ReadMe (for oclFFT tuning for example) . . Hi Raistmer . . I have added -spike_FFT_threshhold 2048 -tune 1 2 1 16 to the command line (from the examples given in the doc file). I have not tried anything from the oclFFT series because I have no idea what "global radix", "local radix" or any of the other terms relate to. Stephen |
©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.