Message boards :
Number crunching :
SETI applications for NVIDIA GPU improvement - how you can help
Message board moderation
Previous · 1 . . . 5 · 6 · 7 · 8 · 9 · 10 · 11 . . . 14 · Next
Author | Message |
---|---|
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
New anticipated default for kernel target time is 60ms. Try to run (both ATi and NV) with -tt 60 for some time and report if GUI lags too high for your particular host. SETI apps news We're not gonna fight them. We're gonna transcend them. |
Mike Send message Joined: 17 Feb 01 Posts: 34362 Credit: 79,922,639 RAC: 80 |
See my bench and coment at Lunatics. With each crime and every kindness we birth our future. |
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
See my bench and coment at Lunatics. Done with request for further testing :) SETI apps news We're not gonna fight them. We're gonna transcend them. |
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
Here are new rev builds, please test https://cloud.mail.ru/public/M7kP/4f6CqSGhB SETI apps news We're not gonna fight them. We're gonna transcend them. |
Zalster Send message Joined: 27 May 99 Posts: 5517 Credit: 528,817,460 RAC: 242 |
Is there anything in particular we are looking for? |
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
Is there anything in particular we are looking for? just general performance and usability. It's new RC builds so if all OK they will go to main unchanged. SETI apps news We're not gonna fight them. We're gonna transcend them. |
Zalster Send message Joined: 27 May 99 Posts: 5517 Credit: 528,817,460 RAC: 242 |
28 Computation errors right away. Not sure if I caused it or if it was from the new app. |
Zalster Send message Joined: 27 May 99 Posts: 5517 Credit: 528,817,460 RAC: 242 |
28 Computation errors right away. http://setiathome.berkeley.edu/result.php?resultid=5133156915 http://setiathome.berkeley.edu/result.php?resultid=5133156917 http://setiathome.berkeley.edu/result.php?resultid=5133151018 just a few |
Richard Haselgrove Send message Joined: 4 Jul 99 Posts: 14676 Credit: 200,643,578 RAC: 874 |
Thanks for the warning - I suspended all tasks, and let them loose one at a time. Resuming a task that had been running under r3500: http://setiathome.berkeley.edu/result.php?resultid=5133129501 Starting a new task from scratch: http://setiathome.berkeley.edu/result.php?resultid=5133135420 Both went through the wisdom and CL compiler stages OK, but crashed (like Zalster's) with a memory access exception as soon as the real work started. That sounds like a program problem to me. NV version, running on device 1, GTX 750 Ti Edit - observation while cleaning BIN and Wisdom files prior to reverting to r3500. Machine has an intel_gpu in addition to the NV cards - but I haven't updated the intel_gpu app - it's still at r3430, and inactive (no tasks). But the NV apps create r3500_IntelRCoreTMi54690CPU350GHz_x86.wisdom r3522_IntelRCoreTMi54690CPU350GHz_x86.wisdom Harmless, but curious. Any suggestion why? BOINC's OpenCL detections are 02/09/2016 18:07:59 | | OpenCL: NVIDIA GPU 0: GeForce GTX 970 (driver version 350.12, device version OpenCL 1.2 CUDA, 4096MB, 3839MB available, 4087 GFLOPS peak) 02/09/2016 18:07:59 | | OpenCL: NVIDIA GPU 1: GeForce GTX 750 Ti (driver version 350.12, device version OpenCL 1.2 CUDA, 2048MB, 1967MB available, 1639 GFLOPS peak) 02/09/2016 18:07:59 | | OpenCL: Intel GPU 0: Intel(R) HD Graphics 4600 (driver version 10.18.10.3621, device version OpenCL 1.2, 1298MB, 1298MB available, 192 GFLOPS peak) 02/09/2016 18:07:59 | | OpenCL CPU: Intel(R) Core(TM) i5-4690 CPU @ 3.50GHz (OpenCL driver vendor: Intel(R) Corporation, driver version 3.0.1.10878, device version OpenCL 1.2 (Build 76413)) Edit 2 - r3500 has been generating that CPU wisdom file all along, timestamped at the start of every task. Just hadn't bothered to look for it before. It's 3 KB, 27 lines. |
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
thanks at least one bug fixed new rev in build SETI apps news We're not gonna fight them. We're gonna transcend them. |
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
Sure, app uses FFTW lib too. SETI apps news We're not gonna fight them. We're gonna transcend them. |
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
https://cloud.mail.ru/public/2F9e/1xm2varnV SETI apps news We're not gonna fight them. We're gonna transcend them. |
Zalster Send message Joined: 27 May 99 Posts: 5517 Credit: 528,817,460 RAC: 242 |
Ok, so far so good. Stderr is showing the correct GPUs being used http://setiathome.berkeley.edu/result.php?resultid=5133497505 http://setiathome.berkeley.edu/result.php?resultid=5133485089 http://setiathome.berkeley.edu/result.php?resultid=5133491214 http://setiathome.berkeley.edu/result.php?resultid=5133491133 |
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
Ok, so far so good. Fftlength=8,pass=3:Tune: sum=5539.19(ms); min=3.181(ms); max=757(ms); mean=240.8(ms); s_mean=208.8; sleep=210(ms); delta=2602; N=23; usual Fftlength=8,pass=4:Tune: sum=3912.48(ms); min=3.409(ms); max=330.2(ms); mean=170.1(ms); s_mean=138.9; sleep=135(ms); delta=2340; N=23; usual Fftlength=8,pass=5:Tune: sum=2917.61(ms); min=3.404(ms); max=250.4(ms); mean=126.9(ms); s_mean=104.9; sleep=105(ms); delta=2077; N=23; usual Do you see any lags? If not consider to add -tt 1500 to effectively disable kernel size tuning. currently you provide period_iterations_num 1, but kernel size exceed default 15ms so adaptation to 15ms-long kernel starts and effective iterations num starts to increase too. SETI apps news We're not gonna fight them. We're gonna transcend them. |
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
And another update: https://cloud.mail.ru/public/LFuj/uqJQnKiMc Hope this one will go on beta servers later. Version to 8.18 now be warned. SETI apps news We're not gonna fight them. We're gonna transcend them. |
Zalster Send message Joined: 27 May 99 Posts: 5517 Credit: 528,817,460 RAC: 242 |
Version to 8.18 now be warned. Wild guess this is actually more for beta identification as we are still 8.12 here at main? I'll test it here in a bit. Some early morning chores here (8 hours behind you Raistmer) still before I can sit down and play with this |
Zalster Send message Joined: 27 May 99 Posts: 5517 Credit: 528,817,460 RAC: 242 |
Ok, so far so good. No lags, I have not tried -tt1500 but I can try with new build r3525. I'll run it with original commandline for a little while, then with the changes you suggested. I'll also decrease the number of instances to 1 work unit per card. These were with 3 per card. |
Zalster Send message Joined: 27 May 99 Posts: 5517 Credit: 528,817,460 RAC: 242 |
And another update: https://cloud.mail.ru/public/LFuj/uqJQnKiMc 1)VHAR http://setiathome.berkeley.edu/result.php?resultid=5134840328 WU true angle range is : 12.356051 Run time 2 min 24 sec CPU time 12 sec 2)VLAR http://setiathome.berkeley.edu/result.php?resultid=5134840293 WU true angle range is : 0.009254 Run time 9 min 10 sec CPU time 9 min 7 sec http://setiathome.berkeley.edu/result.php?resultid=5134840099 WU true angle range is : 0.011647 Run time 9 min 8 sec CPU time 9 min 5 sec 3) Normal AR http://setiathome.berkeley.edu/result.php?resultid=5134800057 WU true angle range is : 0.416899 Run time 4 min 19 sec CPU time 4 min 16 sec http://setiathome.berkeley.edu/result.php?resultid=5134807062 WU true angle range is : 0.416928 Run time 4 min 20 sec CPU time 4 min 17 sec With original commandlines 1) VHAR http://setiathome.berkeley.edu/result.php?resultid=5134821117 WU true angle range is : 2.086420 Run time 2 min 23 sec CPU time 26 sec http://setiathome.berkeley.edu/result.php?resultid=5134821116 WU true angle range is : 2.043501 Run time 2 min 25 sec CPU time 19 sec 2) VLARs http://setiathome.berkeley.edu/result.php?resultid=5134766659 WU true angle range is : 0.022368 Run time 7 min 46 sec CPU time 7 min 44 sec http://setiathome.berkeley.edu/result.php?resultid=5134820892 WU true angle range is : 0.015644 Run time 8 min 5 sec CPU time 8 min 3 sec http://setiathome.berkeley.edu/result.php?resultid=5134820937 WU true angle range is : 0.015644 Run time 8 min 4 sec CPU time 8 min 2 sec 3) Normal AR http://setiathome.berkeley.edu/result.php?resultid=5134766599 WU true angle range is : 0.416902 Run time 4 min 19 sec CPU time 4 min 17 sec http://setiathome.berkeley.edu/result.php?resultid=5134766724 WU true angle range is : 0.432514 Run time 4 min 6 sec CPU time 2 min 38 sec With -tt 1500 no -period_iterations_num 1) VHAR http://setiathome.berkeley.edu/result.php?resultid=5134813343 WU true angle range is : 97.183501 Run time 2 min 24 sec CPU time 7 sec 2) VLARs http://setiathome.berkeley.edu/result.php?resultid=5134840108 WU true angle range is : 0.011647 Run time 7 min 26 sec CPU time 7 min 24 sec http://setiathome.berkeley.edu/result.php?resultid=5134840018 WU true angle range is : 0.011647 Run time 7 min 31 sec CPU time 7 min 29 sec 3) Normal AR http://setiathome.berkeley.edu/result.php?resultid=5134821120 WU true angle range is : 0.447831 Run time 3 min 58 sec CPU time 2 min 54 sec http://setiathome.berkeley.edu/result.php?resultid=5134807348 WU true angle range is : 0.416928 Run time 4 min 17 sec CPU time 3 min 2 sec single work units seem faster with the -tt 1500 than the -period_iterations_num 1 If you want times I can post but I figured you would want to see single runs Edit.. Looking at 3 at a time -tt 1500 http://setiathome.berkeley.edu/result.php?resultid=5134840207[url] WU true angle range is : 0.006972 Run time 21 min 25 sec CPU time 18 min 36 sec http://setiathome.berkeley.edu/result.php?resultid=5134840104 WU true angle range is : 0.011647 Run time 21 min 30 sec CPU time 18 min 48 sec Original commandline http://setiathome.berkeley.edu/result.php?resultid=5134487699 WU true angle range is : 0.024645 Run time 22 min 32 sec CPU time 20 min 29 sec http://setiathome.berkeley.edu/result.php?resultid=5134487700 WU true angle range is : 0.009254 Run time 20 min 50 sec CPU time 19 min 27 sec So not really seeing any advantage to -tt 1500 when running more than 1 work unit per GPU |
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
To get max you should specify both -period_iterations_num 1 and -tt 1500 First one set initial increase and second ensures that app will not change it. SETI apps news We're not gonna fight them. We're gonna transcend them. |
Zalster Send message Joined: 27 May 99 Posts: 5517 Credit: 528,817,460 RAC: 242 |
Ok, I will try it with both, thanks edit. Any particular order in the commandline? -period_iterations_num then -tt or -tt then -period_iterations_num ? |
©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.