Message boards :
Number crunching :
SETI@home v8.12 Windows GPU applications support thread
Message board moderation
Previous · 1 . . . 9 · 10 · 11 · 12 · 13 · 14 · 15 . . . 17 · Next
Author | Message |
---|---|
BilBg Send message Joined: 27 May 07 Posts: 3720 Credit: 9,385,827 RAC: 0 |
Where have you found tasks run on a different GTX 560 ... GTX 560 Ti was notorious to produce false signals (wrong computations) because of too high clock and/or too low voltage (as set by default in the video card factory). I don't know if that is the same/similar for GTX 560 Your GPU seems to run at "Max clock frequency: 1660Mhz" (this is "shader/CUDA Cores" clock): http://setiathome.berkeley.edu/result.php?resultid=5104724560 ... which looks within spec: http://www.geforce.com/hardware/desktop-gpus/geforce-gtx-560/specifications http://www.geforce.com/hardware/desktop-gpus/geforce-gtx-560ti/specifications But nevertheless you can try (for test) to (temporarily) Underclock the GPU and video memory and/or raise the voltage a little to see if that eliminates/decreases # of "Validation inconclusive"/"Invalid". e.g. by MSI Afterburner https://gaming.msi.com/features/afterburner https://www.guru3d.com/files-details/afterburner-download-msi.html http://www.majorgeeks.com/files/details/msi_afterburner.html  - ALF - "Find out what you don't do well ..... then don't do it!" :)  |
robertmiles Send message Joined: 16 Jan 12 Posts: 213 Credit: 4,117,756 RAC: 6 |
Since MOST of the tasks are purged that fast, not enough are left for a proper determination of just how many r3500 tasks were invalid on my GTX 560, so I don't see any point in continuing the discussion of exactly how many there were. I don't have enough free time to inspect my list of tasks every 24 hours, so I usually don't. This means I don't necessarily see all the tasks marked as invalid. I'll try to do a little tuning once a week, but I don't have enough free time to do it every week. |
robertmiles Send message Joined: 16 Jan 12 Posts: 213 Credit: 4,117,756 RAC: 6 |
Where have you found tasks run on a different GTX 560 ... I've never changed those settings from what was set at the factory. I'll add those suggestions to my list of thing to try when I have time, but I don't expect to find time to try more than one change a week. |
robertmiles Send message Joined: 16 Jan 12 Posts: 213 Credit: 4,117,756 RAC: 6 |
I'll let you look for those other hosts - I don't have enough free time to do it. No tuning at all so far. |
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
That's why testing better to conduct on SETI beta. SETI apps news We're not gonna fight them. We're gonna transcend them. |
robertmiles Send message Joined: 16 Jan 12 Posts: 213 Credit: 4,117,756 RAC: 6 |
I've never seen SETI beta. |
robertmiles Send message Joined: 16 Jan 12 Posts: 213 Credit: 4,117,756 RAC: 6 |
[snip] It is the 560. I finally have a little free time and will try to install this change tonight. If I'm able to do so, I'll enable tasks for a few more days. One core shared between GPU tasks and work from the console, so PARTLY freed. |
Kiska Send message Joined: 31 Mar 12 Posts: 302 Credit: 3,067,762 RAC: 0 |
Here is a link to it. You have to register for an account as it is separate from the main project |
Stephen "Heretic" Send message Joined: 20 Sep 12 Posts: 5557 Credit: 192,787,363 RAC: 628 |
. . Hi Raistmer, . . Now that WOW has finished I have taken the bit and done an install of Lunatics Beta(4) selecting SoG r3500 for the GPU. . . As usual the install went without a hitch and SoG in happily crunching the tasks still identified as CUDA50. . . The command line file remains default (ie empty) yet while there is no SoG section in app_config.xml it appears to have followed the settings in the CUDA50 section of that file as it is running 2 WUs on the GPU. . . This affords a ready comparison of the two at the same level. SoG is achieving slightly higher GPU utilisation (up from around 90% to over 95%) and a corresponding increase in GPU temp of about 5 degrees. But there must be an inherent processing advantage with CUDA50 as the run times for normal AR Arecibo work has increased by about 25% (on a sample base of 10/12 tasks). I have now queued an equal number of guppi tasks to see how they fair. Under CUDA50 normAR Arecibo tasks were 60 mins +/- 3 mins, under SoG they are 76 mins +/- 3 mins. Guppi tasks were less consistent under CUDA50 taking anywhere from 120 to 150 mins (still runing two WUs) so I am interested to see if that changes significantly and in which direction. . . You wanted feedback on how r3500 performs on low end NV GPUs. With a CC of 3.5 and CU of only 2 the GT730 (mine has 2GB GDDR5 ram) while high end of low order GPUs is still low order. It is at your disposal for the next week and I will happily try any configuration you wish and report as best I can. . . If I have no response before hand the next step will be to add a SoG section to app_config.xml and set it for single WU execution then try the command line options you have mentioned in the forum (if I can find the relevant messages again that is). . . Have fun :) . . . [edit] On another note I moved half a dozen Arecibo VLAR tasks into the GPU cache on the i5 (GTX950) and they are running quite well as you expected. FWIW runtimes are considerably longer than guppis (on GPU guppis approx 46 mins VLAR approx 80 mins, on the CPU guppis approx 45 mins and VLAR 70 mins) so unless desperate I think I will still leave Arecibo VLAR work for the CPU. For contrast norm AR Arecibo tasks on CPU run for approx 70 to 80 mins but on GPU about 35 mins. . . |
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
But there must be an inherent processing advantage with CUDA50 as the run times for normal AR Arecibo work has increased by about 25% (on a sample base of 10/12 tasks). When I last checked with pre-FERMI GSO9600 and GTX3260 there was pretty big advantage in FFT performance. Own NV cuFFT performed much better that oclFFT on those GPUs. For later GPU models I did not done such profiling so far (still have only taken for some time from friend GT720, other donated GPUs still have to come). So, SETI MB own code should overcome this CUDA vs OpenCL NV advantage. On pre-FERMI GPUs this was not possible so far so I even started CUDA back-port some time ago just to be able to use cuFFT instead of oclFFT. That project quite abandoned for now due to lack of time. Try to add -tt 60 to command line and see how GUI lags change if any. 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 |
But there must be an inherent processing advantage with CUDA50 as the run times for normal AR Arecibo work has increased by about 25% (on a sample base of 10/12 tasks). . . If I understand correctly I should add the cuFFT command and -tt 60 to the command line. As it is, I am not noticing much lag but I will play with the values for -tt and see where lag becomes noticable. I will look in the docs for information on using cuFFT. . . It will be interesting. . . Stephen |
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
cuFFT is own nVidia FFT library available only on CUDA. For OpenCL app I use oclFFT instead. 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 |
cuFFT is own nVidia FFT library available only on CUDA. For OpenCL app I use oclFFT instead. . . OK got it, use oclFFT. Glad to hear that because I could not find cuFFT in the docs ... :) Stephen . . {edit} should I still use -sbs settings or -use_sleep or any other commands as well? |
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
As was mentioned already I look for best defaults, not for best tuning for particular host. New default will be -tt 60 -high_prec_timer -use_sleep for NV and -tt 60 for ATi. That options should be tested. SETI apps news We're not gonna fight them. We're gonna transcend them. |
robertmiles Send message Joined: 16 Jan 12 Posts: 213 Credit: 4,117,756 RAC: 6 |
Has the r3500 testing sucked the supply of Windows GPU workunits dry? A few minutes ago, neither of my computers had any GPU workunits at all, with only SETI@home enabled. Telling both of them to update SETI@home did nothing, so I decided it was time to enable a different project offering GPU workunits. I'm now switching to testing the new default for NV. Do I also need to do any testing I skipped? |
Stephen "Heretic" Send message Joined: 20 Sep 12 Posts: 5557 Credit: 192,787,363 RAC: 628 |
. . I added the -tt 60 and this increased the GPU activity but I did not notice any lag or hesitation when typing and such. Runtimes for Arecibo normAR tasks were typically 46 mins (Cuda50 typically 34 to 36 mins). I increased the -tt value to 120, this improved GPU utilisation even further and runtimes down to approx 42 mins but noticing a little hesitation/stickiness on BOINC manager's slider controls, otherwise typing seems mostly unaffected. . . I will add those commands to the command line to be sure they are running, should I use -use_sleep_ex of 1? . |
Stephen "Heretic" Send message Joined: 20 Sep 12 Posts: 5557 Credit: 192,787,363 RAC: 628 |
Has the r3500 testing sucked the supply of Windows GPU workunits dry? . . There were problems with the splitters last week but now WUs seem to be flowing OK. The project has been down for maintenance in the last 12 hours (started approx 9 hours ago not sure when it finished but has been back up for at least 3 hours) so perhaps you were trying during that time. . . I would give it another try now. It is an often recommended practice to make sure your WU cache is filled before maintenance begins. . . Stephen . |
robertmiles Send message Joined: 16 Jan 12 Posts: 213 Credit: 4,117,756 RAC: 6 |
Has the r3500 testing sucked the supply of Windows GPU workunits dry? Works if you get enough advance notice of the maintenance. I didn't. One computer automatically got more; the other will take longer to be ready for more. |
Grant (SSSF) Send message Joined: 19 Aug 99 Posts: 13757 Credit: 208,696,464 RAC: 304 |
Works if you get enough advance notice of the maintenance. I didn't. It happens every week. It is announced every week on the main page as News. Grant Darwin NT |
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
-use_sleep has same action as -use_sleep_ex 1 SETI apps news We're not gonna fight them. We're gonna transcend them. |
©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.