SETI applications for NVIDIA GPU improvement

Author	Message
Zalster Volunteer tester Send message Joined: 27 May 99 Posts: 5517 Credit: 528,817,460 RAC: 242	Message 1803690 - Posted: 20 Jul 2016, 14:46:47 UTC - in response to Message 1803627. Last modified: 20 Jul 2016, 14:47:11 UTC Came across this while checking r3486 against r3430 It looks like OpenCl Guppi slow down just as cuda work units when paired with an AP on the same device. In this case there were 2 APs and 1 Guppi on the same device. Thought you might want this information AP http://setiathome.berkeley.edu/result.php?resultid=5051041254 Running on device number: 0 Received 20 Jul 2016, 13:45:57 UTC Server state Over Outcome Success Client state Done Exit status 0 (0x0) Computer ID 7565600 Run time 30 min CPU time 29 min 50 sec AP http://setiathome.berkeley.edu/result.php?resultid=5051041221 Running on device number: 0 Received 20 Jul 2016, 13:45:57 UTC Server state Over Outcome Success Client state Done Exit status 0 (0x0) Computer ID 7565600 Run time 30 min 8 sec CPU time 29 min 56 sec Guppi http://setiathome.berkeley.edu/result.php?resultid=5050524994 Running on device number: 0 Received 20 Jul 2016, 13:51:07 UTC Server state Over Outcome Success Client state Done Exit status 0 (0x0) Computer ID 7565600 Run time 28 min 41 sec CPU time 24 min 22 sec As a side r3430 is still about 60-90sec faster than r3486 when doing 3 at a time ID: 1803690 · Reply Quote

Raistmer Volunteer developer Volunteer tester Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121	Message 1803691 - Posted: 20 Jul 2016, 14:48:51 UTC - in response to Message 1803690. As a side r3430 is still about 60-90sec faster than r3486 when doing 3 at a time What tuning lines were used in this comparison? SETI apps news We're not gonna fight them. We're gonna transcend them. ID: 1803691 · Reply Quote

Zalster Volunteer tester Send message Joined: 27 May 99 Posts: 5517 Credit: 528,817,460 RAC: 242	Message 1803693 - Posted: 20 Jul 2016, 14:55:44 UTC - in response to Message 1803691. Yes, was using my commandlines on both machines for this comparison ID: 1803693 · Reply Quote

Raistmer Volunteer developer Volunteer tester Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121	Message 1803729 - Posted: 20 Jul 2016, 16:44:48 UTC - in response to Message 1803693. Yes, was using my commandlines on both machines for this comparison What command lines, please list them. SETI apps news We're not gonna fight them. We're gonna transcend them. ID: 1803729 · Reply Quote

Stephen "Heretic" Volunteer tester Send message Joined: 20 Sep 12 Posts: 5557 Credit: 192,787,363 RAC: 628	Message 1804802 - Posted: 25 Jul 2016, 11:58:15 UTC - in response to Message 1803533. Please test in single task per GPU mode with and w/o -use_sleep for comparison: https://cloud.mail.ru/public/6wp7/cgfuAXmnc Is this a new version or still part of the older r3486Final? revision the same but binary is new. . . I am working up the courage to try and implement this version. Am I right to presume that it runs the standard MB8.12 WUs? ID: 1804802 · Reply Quote

Raistmer Volunteer developer Volunteer tester Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121	Message 1805166 - Posted: 27 Jul 2016, 18:37:35 UTC - in response to Message 1803476. Another test to do by volunteers: On beta (or in anonymous platform mode on main) take 8.16 OpenCL app or later and add -tt F parameter to command line. increase F value (default is 15 that means 15 ms target time for partial PulseFind kernel) and watch for host usability (that is, GUI lags, missing letters at typing and so on). At what F value lags appear? From performance point of view it's better to have longer kernels but this can result in GUI lags. This testing needed to establish best possible default value for unattended run. EDIT: describe your host config (preferably with link to host on beta) along with report, please. I would like to remind that w/o feedback about usability there will be no any corrections done and chance to get app working nice out of the box for your particular host will be missed. Please check usability of 8.16+ builds for your particular host and report back. SETI apps news We're not gonna fight them. We're gonna transcend them. ID: 1805166 · Reply Quote

Harri Liljeroos Send message Joined: 29 May 99 Posts: 4635 Credit: 85,281,665 RAC: 126	Message 1805206 - Posted: 27 Jul 2016, 21:46:08 UTC - in response to Message 1805166. Well my experience without the -use_sleep parameter was a driver restart within 5 minutes of starting the application. So I was forced to go back to -use_sleep. The application does use a lot less CPU compared to previous versions. I am currently running with -period_iterations_num 5 with no lag on this host: http://setiathome.berkeley.edu/results.php?hostid=7043787, decreasing it to 3 and lags will appear. The host has two GPUs: a GTX650Ti and a GTX970 both running one seti task at a time with three CPU cores reserved for GPU feeding. The host's GPUs do also Einstein WUs one or two at a time depending on how BOINC is scheduling them (seti share is 45 and einstein is 25). ID: 1805206 · Reply Quote

Raistmer Volunteer developer Volunteer tester Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121	Message 1805214 - Posted: 27 Jul 2016, 22:31:07 UTC - in response to Message 1805206. And what about running with defaults? SETI apps news We're not gonna fight them. We're gonna transcend them. ID: 1805214 · Reply Quote

Harri Liljeroos Send message Joined: 29 May 99 Posts: 4635 Credit: 85,281,665 RAC: 126	Message 1805323 - Posted: 28 Jul 2016, 12:37:43 UTC - in response to Message 1805214. Last modified: 28 Jul 2016, 12:41:18 UTC And what about running with defaults? Today I have started running with default settings, so no commandline and lately also without app_config. I am positively surprised how well the host is behaving. The load on GPUs is of course fluctuating as there are no free CPU cores to feed them constantly but I have no big lags or driver restarts. On CPU there is running 6 setiathome_v8 tasks and 2 CPDN wah2_eu25 tasks so CPU load is constantly at 100%. I can also use the computer for email, surfing the net and I could view the new Seti video from Youtube without a problem (Firefox HW-acceleration is turned off). I am using the r3486 release with date 19.7.2016 on the exe. Here are a few WUs I have crunched so far with default settings: http://setiathome.berkeley.edu/result.php?resultid=5063178279 a Guppi VLAR on GTX970. http://setiathome.berkeley.edu/result.php?resultid=5063162557 normal Arecibo WU on GTX970. http://setiathome.berkeley.edu/result.php?resultid=5063120898 normal Arecibo WU on GTX650Ti. http://setiathome.berkeley.edu/result.php?resultid=5063178278 a Guppi VLAR on GTX650Ti. So good job Raistmer on this application. (edit)Next I try and free one CPU core and see what that does.(/edit) ID: 1805323 · Reply Quote

Raistmer Volunteer developer Volunteer tester Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121	Message 1805431 - Posted: 28 Jul 2016, 21:53:48 UTC - in response to Message 1805323. Thanks for report. Good that no usability problems at defaults. SETI apps news We're not gonna fight them. We're gonna transcend them. ID: 1805431 · Reply Quote

Harri Liljeroos Send message Joined: 29 May 99 Posts: 4635 Credit: 85,281,665 RAC: 126	Message 1805554 - Posted: 29 Jul 2016, 9:26:10 UTC The test with freeing a CPU core for feeding the GPUs resulted the already known conclusion: one free CPU core is not enough for two GPUs to maximize the GPU load. For best result you should free one CPU core for each GPU (while running one WU/GPU at a time). This is for my host, but YMMV. Now I'm back to using my old app_config and commnadline parameters. ID: 1805554 · Reply Quote

Raistmer Volunteer developer Volunteer tester Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121	Message 1805560 - Posted: 29 Jul 2016, 10:08:24 UTC - in response to Message 1805554. The test with freeing a CPU core for feeding the GPUs resulted the already known conclusion: one free CPU core is not enough for two GPUs to maximize the GPU load. For best result you should free one CPU core for each GPU (while running one WU/GPU at a time). This is for my host, but YMMV. Now I'm back to using my old app_config and commnadline parameters. I would recommend to try also -hp -cpu_lock addition to your tuning line to improve GPU load with busy CPU. SETI apps news We're not gonna fight them. We're gonna transcend them. ID: 1805560 · Reply Quote

Harri Liljeroos Send message Joined: 29 May 99 Posts: 4635 Credit: 85,281,665 RAC: 126	Message 1805568 - Posted: 29 Jul 2016, 11:24:22 UTC - in response to Message 1805560. Last modified: 29 Jul 2016, 11:24:52 UTC The test with freeing a CPU core for feeding the GPUs resulted the already known conclusion: one free CPU core is not enough for two GPUs to maximize the GPU load. For best result you should free one CPU core for each GPU (while running one WU/GPU at a time). This is for my host, but YMMV. Now I'm back to using my old app_config and commnadline parameters. I would recommend to try also -hp -cpu_lock addition to your tuning line to improve GPU load with busy CPU. The -hp was there already but I'm adding the -cpu_lock now. So new commandline is: -sbs 1024 -use_sleep -hp -period_iterations_num 5 -instances_per_device 1 -spike_fft_thresh 4096 -tune 1 64 1 4 -oclfft_tune_gr 256 -oclfft_tune_lr 16 -oclfft_tune_wg 256 -oclfft_tune_ls 512 -oclfft_tune_bn 32 -oclfft_tune_cw 32 -tt 75 -cpu_lock The -sbs 1024 applies to GTX970 only as the application seems to limit that value to 25% of GPU's memory. So GTX 650Ti only uses -sbs 256. ID: 1805568 · Reply Quote

Raistmer Volunteer developer Volunteer tester Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121	Message 1805601 - Posted: 29 Jul 2016, 15:33:40 UTC New set of builds (r3500) available here: https://cloud.mail.ru/public/LJ8s/c3WyRR8ip Please test SETI apps news We're not gonna fight them. We're gonna transcend them. ID: 1805601 · Reply Quote

Harri Liljeroos Send message Joined: 29 May 99 Posts: 4635 Credit: 85,281,665 RAC: 126	Message 1805930 - Posted: 30 Jul 2016, 21:59:39 UTC - in response to Message 1805601. New set of builds (r3500) available here: https://cloud.mail.ru/public/LJ8s/c3WyRR8ip Please test I don't see much difference to the previous version. Slightly more lags but not too bad. When running with defaults same situation. When I freed two CPU cores and tried with a minimal commandline: -sbs 1024 -hp -instances_per_device 1 -cpu_lock I got a driver restart. Here's the WU where it happened: http://setiathome.berkeley.edu/result.php?resultid=5065787291 ID: 1805930 · Reply Quote

tazzduke Volunteer tester Send message Joined: 15 Sep 07 Posts: 190 Credit: 28,269,068 RAC: 5	Message 1805990 - Posted: 31 Jul 2016, 4:35:52 UTC Greetings Raistmer Have got brought back online two Crunchers Cruncher 1 - LGA775 Q6600 Win 7 x64 Boinc 7.6.22 with HD7870 and GTX 760 Cruncher 2 - LGA775 Q9300 Win 7 x64 Boinc 7.6.22 with GTX 770 Running build r3500 with defaults (no cmd line switches) Will run this for awhile, noticing no screen lag on both crunchers. By the way only doing GPU crunching on these machines. When I get time, I will try and post work unit details. Regards ID: 1805990 · Reply Quote

Zalster Volunteer tester Send message Joined: 27 May 99 Posts: 5517 Credit: 528,817,460 RAC: 242	Message 1806195 - Posted: 31 Jul 2016, 23:16:11 UTC r3500 A. CPU consumption 1. No Sleep command WU true angle range is : 1.477986 http://setiathome.berkeley.edu/result.php?resultid=5070945053 Run time 1 min 6 sec CPU time 10 sec WU true angle range is : 0.415260 http://setiathome.berkeley.edu/result.php?resultid=5070939131 Run time 4 min 28 sec CPU time 4 min 24 sec WU true angle range is : 0.024646 http://setiathome.berkeley.edu/result.php?resultid=5070945029 Run time 8 min 3 sec CPU time 7 min 51 sec WU true angle range is : 0.009960 (GUPPI) http://setiathome.berkeley.edu/result.php?resultid=5070939038 Run time 7 min 40 sec CPU time 7 min 36 sec 2. No sleep command but -tt 1500 WU true angle range is : 1.477986 http://setiathome.berkeley.edu/result.php?resultid=5070945080 Run time 2 min 30 sec CPU time 23 sec WU true angle range is : 0.415318 http://setiathome.berkeley.edu/result.php?resultid=5070945010 Run time 4 min 26 sec CPU time 4 min 4 sec WU true angle range is : 0.011214 (GUPPI) http://setiathome.berkeley.edu/result.php?resultid=5070952849 Run time 7 min 41 sec CPU time 7 min 38 sec 3. -use_sleep WU true angle range is : 1.477986 http://setiathome.berkeley.edu/result.php?resultid=5070945148 Run time 2 min 34 sec CPU time 9 sec WU true angle range is : 0.415260 http://setiathome.berkeley.edu/result.php?resultid=5070939117 Run time 6 min 19 sec CPU time 1 min 1 sec WU true angle range is : 0.330733 http://setiathome.berkeley.edu/result.php?resultid=5070938858 Run time 7 min 31 sec CPU time 1 min 9 sec WU true angle range is : 0.011214 (GUPPI) http://setiathome.berkeley.edu/result.php?resultid=5070939149 Run time 11 min 35 sec CPU time 2 min 19 sec 4. -use_sleep -high_prec_timer WU true angle range is : 1.484248 http://setiathome.berkeley.edu/result.php?resultid=5070938888 Run time 2 min 33 sec CPU time 10 sec WU true angle range is : 1.484248 http://setiathome.berkeley.edu/result.php?resultid=5070938911 Run time 2 min 32 sec CPU time 25 sec WU true angle range is : 0.415260 http://setiathome.berkeley.edu/result.php?resultid=5070939125 Run time 4 min 35 sec CPU time 1 min 49 sec WU true angle range is : 0.012603 (GUPPI) http://setiathome.berkeley.edu/result.php?resultid=5070939088 Run time 7 min 53 sec CPU time 3 min 8 sec WU true angle range is : 0.012603 (GUPPI) http://setiathome.berkeley.edu/result.php?resultid=5070939088 Run time 7 min 53 sec CPU time 3 min 8 sec 5. -use_sleep -tt 1500 WU true angle range is : 8.804756 http://setiathome.berkeley.edu/result.php?resultid=5070939147 Run time 2 min 26 sec CPU time 7 sec WU true angle range is : 1.484248 http://setiathome.berkeley.edu/result.php?resultid=5070938887 Run time 2 min 35 sec CPU time 9 sec WU true angle range is : 0.330733 http://setiathome.berkeley.edu/result.php?resultid=5070939071 Run time 5 min 4 sec CPU time 4 min 17 sec WU true angle range is : 0.330733 http://setiathome.berkeley.edu/result.php?resultid=5070938786 Run time 7 min 31 sec CPU time 1 min 7 sec WU true angle range is : 0.032335 (GUPPI) http://setiathome.berkeley.edu/result.php?resultid=5070945042 Run time 10 min 59 sec CPU time 2 min Aside Oddball work unit blc 3 prior attempts ended in error. This one complete in under 12 sec WU true angle range is : 0.027435 http://setiathome.berkeley.edu/result.php?resultid=5070939086 Run time 11 sec CPU time 8 sec ID: 1806195 · Reply Quote

Richard Haselgrove Volunteer tester Send message Joined: 4 Jul 99 Posts: 14672 Credit: 200,643,578 RAC: 874	Message 1806197 - Posted: 31 Jul 2016, 23:22:03 UTC - in response to Message 1806195. Aside Oddball work unit blc 3 prior attempts ended in error. This one complete in under 12 sec The three prior attempts were all with the stock application with over-zealous sanity checks on the 'found' signals. That's the one we're desperately trying to replace (arguably a premature release) with an application version similar to the one you're testing now. ID: 1806197 · Reply Quote

Zalster Volunteer tester Send message Joined: 27 May 99 Posts: 5517 Credit: 528,817,460 RAC: 242	Message 1806231 - Posted: 1 Aug 2016, 1:11:22 UTC - in response to Message 1806195. Raistmer, what is the difference between -tt 1500 and -high_prec_timer? It looks to me like the -high_prec_timer might be faster ID: 1806231 · Reply Quote

Raistmer Volunteer developer Volunteer tester Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121	Message 1806457 - Posted: 1 Aug 2016, 20:41:05 UTC - in response to Message 1805930. I got a driver restart. Here's the WU where it happened: http://setiathome.berkeley.edu/result.php?resultid=5065787291 Unfortunately result out of reach already. Worth to copy stderr ot do some work at beta. SETI apps news We're not gonna fight them. We're gonna transcend them. ID: 1806457 · Reply Quote

©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.

SETI applications for NVIDIA GPU improvement - how you can help