v8.19 test results

Author	Message
AMDave Volunteer tester Send message Joined: 12 Jan 16 Posts: 38 Credit: 289,647 RAC: 0	Message 59591 - Posted: 17 Sep 2016, 0:56:54 UTC GPU is not OCâ€™d, itâ€™s strictly stock with default settings.Â Â 1 WU/GPU. >Â Â stock Task 24802355 WU true angle range : 0.007977 Run time : 15m 03s CPU time : 14m 42s Task 24801929 WU true angle range : 2.866236 Run time : 06m 30s CPU time : 00m 48s Task 24802557 WU true angle range : 0.007977 Run time : 14m 59s CPU time : 14m 25s Task 24802388 WU true angle range : 0.007977 Run time : 14m 52s CPU time : 14m 37s >Â Â -tt 30 Task 24802346 WU true angle range : 0.007977 Run time : 15m 01s CPU time : 14m 41s Task 24802340 WU true angle range : 2.866236 Run time : 06m 27s CPU time : 00m 49s Task 24802390 WU true angle range : 0.007977 Run time : 14m 58s CPU time : 14m 47s Task 24802221 WU true angle range : 2.866236 Run time : 06m 26s CPU time : 00m 47s >Â Â -tt 60 Task 24802400 WU true angle range : 2.866236 Run time : 06m 30s CPU time : 00m 47s Task 24802331 WU true angle range : 0.007977 Run time : 14m 54s CPU time : 14m 42s Task 24802060 WU true angle range : 0.007977 Run time : 14m 49s CPU time : 14m 41s Task 24801377 WU true angle range : 2.866236 Run time : 06m 27s CPU time : 00m 46s >Â Â -tt 90 Task 24802277 WU true angle range : 0.007977 Run time : 14m 47s CPU time : 14m 36s Task 24801614 WU true angle range : 2.866236 Run time : 06m 27s CPU time : 00m 45s Task 24801967 WU true angle range : 0.007977 Run time : 14m 53s CPU time : 14m 34s Task 24806655 WU true angle range : 1.123006 Run time : 07m 43s CPU time : 02m 40s >Â Â -tt 120 Task 24802402 WU true angle range : 0.007977 Run time : 14m 51s CPU time : 14m 37s Task 24802266 WU true angle range : 2.866236 Run time : 06m 26s CPU time : 00m 45s Task 24802342 WU true angle range : 0.007977 Run time : 14m 41s CPU time : 14m 34s Task 24802528 WU true angle range : 2.866236 Run time : 06m 26s CPU time : 00m 44s >Â Â -tt 180 Task 24802524 WU true angle range : 2.866236 Run time : 06m 25s CPU time : 00m 45s Task 24802590 WU true angle range : 0.007977 Run time : 14m 45s CPU time : 14m 37s Task 24802398 WU true angle range : 0.007977 Run time : 14m 48s CPU time : 14m 34s Task 24802540 WU true angle range : Run time : 06m 27s CPU time : 00m 47s >Â Â -tt 30 -sbs 256 Task 24806696 WU true angle range : 1.123006 Run time : 08m 07s CPU time : 02m 55s Task 24806654 WU true angle range : 0.007806 Run time : 14m 55s CPU time : 14m 44s Task 24806631 WU true angle range : 1.123006 Run time : 08m 09s CPU time : 03m 04s Task 24806437 WU true angle range : 0.007806 Run time : 14m 58s CPU time : 14m 43s NOTES -- VHAR nearly 100 sec > stock (6:30 v 8:10), VLAR virtually identical, GUI lag intermittent >Â Â -tt 60 -sbs 256 Task 24806686 WU true angle range : 1.123006 Run time : 08m 09s CPU time : 03m 02s Task 24806651 WU true angle range : 1.123006 Run time : 08m 07s CPU time : 03m 02s Task 24806741 WU true angle range : 0.007806 Run time : 14m 52s CPU time : 14m 37s Task 24811854 WU true angle range : 0.011707 Run time : 23m 11s CPU time : 23m 43s NOTES -- VHAR nearly 100 sec > stock (6:30 v 8:10), GUI lag frequent VLAR virtually identical, GUI lag more noticeable while surfing web and switching apps >Â Â -use_sleep with -high_prec_timer Task 24802460 WU true angle range : 0.007977 Run time : 15m 39s CPU time : 04m 07s Task 24800827 WU true angle range : 1.189277 Run time : 06m 41s CPU time : 00m 43s Task 24802302 WU true angle range : 0.007977 Run time : 16m 00s CPU time : 04m 02s Task 24802249 WU true angle range : 2.866236 Run time : 06m 29s CPU time : 00m 36s More to follow if need be. ID: 59591 ·

Raistmer Volunteer tester Send message Joined: 18 Aug 05 Posts: 2423 Credit: 15,878,738 RAC: 0	Message 59596 - Posted: 17 Sep 2016, 10:45:15 UTC - in response to Message 59591. Last modified: 17 Sep 2016, 10:54:18 UTC Thanks for report. I would like to get more statistics on VHAR difference you noticed. Let's see here: https://setiweb.ssl.berkeley.edu/beta/result.php?resultid=24806655 It's -tt 90 "stock" (almost stock more precise). You list it in so speaking "fast VHAR" area (slow VHAR area ~8min with comments about slowdown). So I would like to get more statistics between this "fast" and "slow" - are they really different (that is, relate upon tuning line) or it's just big variation in run times for VHAR per se. So, please look for AR >1 tasks, and collect more such tasks for let say "pure stock" (-tt 60), "increased" (-tt 90) and "expanded" (-tt 90 -sbs 256). Lets limit to only 3 categories for now, but with good statistics each. And for the record: Name: GeForce GTX 950 Vendor: NVIDIA Corporation Driver version: 359.00 EDIT: And regarding lags: VLAR virtually identical, GUI lag more noticeable while surfing web and switching apps Would be good to spot what AR of task in crunching was at time of lags. Do you see lags for all ARs? Or only for -sbs 256 + VLAR ? Or maybe reverse, -sbs 256 + VHAR ? W/o direct profiling data this info would help to determine where in processing chain lag could appear. News about SETI opt app releases: https://twitter.com/Raistmer ID: 59596 ·

AMDave Volunteer tester Send message Joined: 12 Jan 16 Posts: 38 Credit: 289,647 RAC: 0	Message 59598 - Posted: 17 Sep 2016, 18:12:49 UTC - in response to Message 59596. I initiated testing based on your posts Message 1809598 Message 1802005 The following were run prior to reading your post. >Â Â -high_perf Task 24811662 WU true angle range : 0.011707 Run time : 22m 58s CPU time : 22m 39s Task 24811633 WU true angle range : 0.011707 Run time : 22m 36s CPU time : 22m 36s Task 24812627 WU true angle range : 1.954970 Run time : 06m 38s CPU time : 01m 06s Task 24812847 WU true angle range : 1.158120 Run time : 06m 37s CPU time : 01m 25s NOTES -- VLAR GUI lag unacceptable.Â Â VHAR GUI lag much better, but not as well as stock. So, please look for AR >1 tasks, and collect more such tasks for let say "pure stock" (-tt 60), "increased" (-tt 90) and "expanded" (-tt 90 -sbs 256). Lets limit to only 3 categories for now, but with good statistics each. Working on.Â Â I'm trying to get 5 WUs for each. EDIT: And regarding lags: VLAR virtually identical, GUI lag more noticeable while surfing web and switching apps Would be good to spot what AR of task in crunching was at time of lags. Do you see lags for all ARs? Or only for -sbs 256 + VLAR ? Or maybe reverse, -sbs 256 + VHAR ? W/o direct profiling data this info would help to determine where in processing chain lag could appear. I only tested 2 VLARs & 2 VHARs with those switches, so my observation is generalized.Â Â Can test more WUs with -tt F -sbs 256 switches if you require. >Â Â stock (no switches) Task 24813261 WU true angle range : 1.158120 Run time : 06m 40s CPU time : 01m 26s Task 24812770 WU true angle range : 1.158120 Run time : 06m 38s CPU time : 01m 26s Task 24813611 WU true angle range : 1.078262 Run time : 08m 10s CPU time : 03m 06s Task 24814483 WU true angle range : 1.078262 Run time : 08m 09s CPU time : 03m 02s Task 24814439 WU true angle range : 1.078262 Run time : 08m 08s CPU time : 03m 05s NOTES -- No GUI lag. ID: 59598 ·

Raistmer Volunteer tester Send message Joined: 18 Aug 05 Posts: 2423 Credit: 15,878,738 RAC: 0	Message 59599 - Posted: 17 Sep 2016, 18:58:02 UTC - in response to Message 59598. Last modified: 17 Sep 2016, 19:01:07 UTC Looks like AR=1.0 ~8min while AR=1.1~6min40s So, AR-dependence here, not tuning line one. Regarding -high_perf - bigger lag is expected indeed. EDIT: note that your prev run (14 min) VLARs were GUPPI ones (BLC*) - they base speed different from Arecibo VLAR (22 min in last test). Worth to compare -high_perf on GUPPI now (regarding lags too). Arecibo VLAR is hardest one regarding lags BTW. News about SETI opt app releases: https://twitter.com/Raistmer ID: 59599 ·

AMDave Volunteer tester Send message Joined: 12 Jan 16 Posts: 38 Credit: 289,647 RAC: 0	Message 59600 - Posted: 17 Sep 2016, 20:01:31 UTC - in response to Message 59599. I may have misinterpreted your earlier post.Â Â Now, I believe you may have wanted -sbs 256 included with the following >Â Â -tt 60 Task 24817351 WU true angle range : 0.886736 Run time : 08m 37s CPU time : 04m 19s Task 24817678 WU true angle range : 0.886736 Run time : 08m 27s CPU time : 04m 45s Task 24824577 WU true angle range : 0.832014 Run time : 08m 37s CPU time : 05m 09s Task 24824685 WU true angle range : 0.832014 Run time : 08m 38s CPU time : 05m 07s Task 24824448 WU true angle range : 0.832014 Run time : 08m 40s CPU time : 05m 06s NOTES -- GUI lag very minimal. Unfortunately, not much variation in AR. >Â Â -tt 90 Task 24824499 WU true angle range : 0.832014 Run time : 08m 38s CPU time : 05m 12s Task 24824564 WU true angle range : 0.832014 Run time : 08m 38s CPU time : 05m 11s Task 24824682 WU true angle range : 0.832014 Run time : 08m 37s CPU time : 05m 10s Task 24824792 WU true angle range : 0.832014 Run time : 08m 27s CPU time : 05m 00s Task 24824587 WU true angle range : 0.832014 Run time : 08m 22s CPU time : 04m 52s NOTES -- GUI lag slightly more noticeable than -tt 60 ID: 59600 ·

AMDave Volunteer tester Send message Joined: 12 Jan 16 Posts: 38 Credit: 289,647 RAC: 0	Message 59601 - Posted: 18 Sep 2016, 0:30:48 UTC Last modified: 18 Sep 2016, 0:32:23 UTC The servers gave only 1 AR. >Â Â -tt 90 -sbs 256 Task 24824801 WU true angle range : 0.832014 Run time : 08m 39s CPU time : 05m 06s Task 24824694 WU true angle range : 0.832014 Run time : 08m 38s CPU time : 05m 03s Task 24824679 WU true angle range : 0.832014 Run time : 08m 40s CPU time : 05m 05s Task 24824618 WU true angle range : 0.832014 Run time : 08m 20s CPU time : 04m 45s WU true angle range : 0.832014 Run time : 08m 28s CPU time : 04m 57s NOTES -- GUI lag slightly more noticeable than -tt 90.Â Â Compared to stock, these switches caused Run time at least 30s longer and CPU time 90s longer.Â Â GUI lag noticeable when surfing web (page loading, scrolling, tab switching) see * below. >Â Â -tt 60 -sbs 256 Task 24824798 WU true angle range : 0.832014 Run time : 08m 42s CPU time : 05m 12s Task 24824795 WU true angle range : 0.832014 Run time : 08m 40s CPU time : 05m 07s Task 24824786 WU true angle range : 0.832014 Run time : 08m 35s CPU time : 04m 58s Task 24824546 WU true angle range : 0.832014 Run time : 08m 40s CPU time : 05m 09s Task 24824723 WU true angle range : 0.832014 Run time : 08m 22s CPU time : 04m 49s NOTES -- GUI lag similar to -tt 90 -sbs 256.Â Â Compared to -tt 90 -sbs 256, these switches caused Run time about the same and CPU time 5s-10s longer.Â Â GUI lag noticeable when surfing web (page loading, scrolling, tab switching) see * below. >Â Â -tt 30 -sbs 256 Task 24824569 WU true angle range : 0.832014 Run time : 08m 30s CPU time : 00m 05s Task 24824561 WU true angle range : 0.832014 Run time : 08m 28s CPU time : 05m 03s Task 24824729 WU true angle range : 0.832014 Run time : 08m 24s CPU time : 04m 52s Task 24824705 WU true angle range : 0.832014 Run time : 08m 38s CPU time : 05m 05s Task 24824773 WU true angle range : 0.832014 Run time : 08m 36s CPU time : 05m 05s NOTES -- GUI lag not as pronounced as -tt 90 -sbs 256, and -tt 60 -sbs 256.Â Â Compared to those, the Run time and the CPU time were roughly similar.Â Â GUI lag less noticeable when surfing web (page loading, scrolling, tab switching) see * below. *Â Â GPU load is usually 98%-99% with only S@h running.Â Â More often than not, I suspend GPU when web surfing or the occasional game. ID: 59601 ·

Raistmer Volunteer tester Send message Joined: 18 Aug 05 Posts: 2423 Credit: 15,878,738 RAC: 0	Message 59607 - Posted: 18 Sep 2016, 7:55:23 UTC - in response to Message 59601. >Â Â -tt 90 -sbs 256[list] WU true angle range : 0.832014 Run time : 08m 28s CPU time : 04m 57s NOTES -- GUI lag slightly more noticeable than -tt 90.Â Â Compared to stock, these switches caused Run time at least 30s longer and CPU time 90s longer.Â With what stock run you do compare exactly? And some analyse of this particular result: Target kernel sequence time set to 90ms Maximum single buffer size set to:256MB Name: GeForce GTX 950 Driver version: 359.00 WU true angle range is : 0.832014 Used GPU device parameters are: Number of compute units: 6 Single buffer allocation size: 256MB Total device global memory: 2048MB max WG size: 1024 local mem type: Real FERMI path used: yes LotOfMem path: yes LowPerformanceGPU path: no HighPerformanceGPU path: no period_iterations_num=50 Fftlength=8,pass=3:Tune: sum=2337.32(ms); min=2.284(ms); max=40.32(ms); mean=23.37(ms); s_mean=30.14; sleep=30(ms); delta=131; N=100; usual Fftlength=8,pass=4:Tune: sum=1683.08(ms); min=2.84(ms); max=29.6(ms); mean=17.91(ms); s_mean= 26; sleep=15(ms); delta=125; N=94; usual Fftlength=8,pass=5:Tune: sum=1319.88(ms); min=1.831(ms); max=24.57(ms); mean=14.83(ms); s_mean=17.73; sleep=15(ms); delta=120; N=89; usual Fftlength=16,pass=3:Tune: sum=1639.9(ms); min=1.023(ms); max=28.26(ms); mean=13.67(ms); s_mean=22.62; sleep=15(ms); delta=135; N=120; usual Fftlength=16,pass=4:Tune: sum=1163.15(ms); min=0.7331(ms); max=19.78(ms); mean=10.2(ms); s_mean=16.4; sleep=15(ms); delta=129; N=114; usual Fftlength=16,pass=5:Tune: sum=961.457(ms); min=1.1(ms); max=17.64(ms); mean=8.986(ms); s_mean=13.14; sleep=15(ms); delta=122; N=107; usual Fftlength=32,pass=3:Tune: sum=1115.22(ms); min=0.3611(ms); max=20.74(ms); mean=9.067(ms); s_mean= 15; sleep=15(ms); delta=130; N=123; usual Fftlength=32,pass=4:Tune: sum=738.087(ms); min=0.3559(ms); max=13.85(ms); mean=6.474(ms); s_mean=10.32; sleep=0(ms); delta=121; N=114; usual Fftlength=32,pass=5:Tune: sum=618.278(ms); min=0.3963(ms); max=12.33(ms); mean=5.778(ms); s_mean=8.444; sleep=0(ms); delta=114; N=107; usual Fftlength=64,pass=3:Tune: sum=2602.49(ms); min=23.31(ms); max=52.25(ms); mean=51.03(ms); s_mean=51.79; sleep=45(ms); delta=1; N=51; high_perf Fftlength=128,pass=3:Tune: sum=2811.65(ms); min=13.5(ms); max=28.9(ms); mean=27.84(ms); s_mean=28.1; sleep=30(ms); delta=1; N=101; high_perf Fftlength=256,pass=3:Tune: sum=2965.75(ms); min=6.452(ms); max=14.97(ms); mean=14.61(ms); s_mean=14.75; sleep=15(ms); delta=1; N=203; high_perf Fftlength=512,pass=3:Tune: sum=1019.88(ms); min=2.404(ms); max=2.709(ms); mean=2.506(ms); s_mean=2.535; sleep=0(ms); delta=1; N=407; usual Fftlength=1024,pass=3:Tune: sum=664.134(ms); min=0.7999(ms); max=0.853(ms); mean=0.8169(ms); s_mean=0.8173; sleep=0(ms); delta=1; N=813; usual As one can see, target kernel time of 90ms can't be reached on this AR for this particular GPU. Longest kernel sequence for PulseFind is: Fftlength=64,pass=3:Tune: sum=2602.49(ms); min=23.31(ms); max=52.25(ms); mean=51.03(ms); s_mean=51.79; sleep=45(ms); delta=1; N=51; high_perf It's quite low value and hardly would cause any lags. That means lags are from some other kernel. I suspect local PulseFind. If you able to go to anonymous platform I would provide special build to check this. But before while you have exactly this AR batch please add -use_sleep to -tt 90 -sbs 256. How sleep addition will influent both run time and lags? News about SETI opt app releases: https://twitter.com/Raistmer ID: 59607 ·

AMDave Volunteer tester Send message Joined: 12 Jan 16 Posts: 38 Credit: 289,647 RAC: 0	Message 59609 - Posted: 18 Sep 2016, 15:37:17 UTC - in response to Message 59607. Last modified: 18 Sep 2016, 15:37:32 UTC >Â Â -tt 90 -sbs 256 WU true angle range : 0.832014 Run time : 08m 28s CPU time : 04m 57s NOTES -- GUI lag slightly more noticeable than -tt 90.Â Â Compared to stock, these switches caused Run time at least 30s longer and CPU time 90s longer.Â With what stock run you do compare exactly? Stock = the initial 4 results from above in Message 59591.Â Â Maybe I should label it "BASELINE". As one can see, target kernel time of 90ms can't be reached on this AR for this particular GPU. Longest kernel sequence for PulseFind is: Fftlength=64,pass=3:Tune: sum=2602.49(ms); min=23.31(ms); max=52.25(ms); mean=51.03(ms); s_mean=51.79; sleep=45(ms); delta=1; N=51; high_perf It's quite low value and hardly would cause any lags. That means lags are from some other kernel. I suspect local PulseFind. [/quote] Forgive me, I don't fully understand the inner complexities yet.Â Â I do read the README.txt files, and I'm in the process of reading your posts on the Lunatics forum, like this. If you able to go to anonymous platform I would provide special build to check this. Yes, but I would need assistance with the app_info or app_config file. But before while you have exactly this AR batch please add -use_sleep to -tt 90 -sbs 256. How sleep addition will influent both run time and lags? > -tt 90 -sbs 256 -use_sleep Task 24824697 WU true angle range : 0.832014 Run time : 09m 09s CPU time : 00m 55s Task 24824558 WU true angle range : 0.832014 Run time : 09m 01s CPU time : 01m 01s Task 24824542 WU true angle range : 0.832014 Run time : 08m 44s CPU time : 01m 02s Task 24832279 WU true angle range : 0.010849Â Â Â (oops, a VLAR snuck in here) Run time : 26m 42s CPU time : 04m 54s Task 24832395 WU true angle range : 3.417951 Run time : 06m 27s CPU time : 00m 18s Task 24832338 WU true angle range : 3.417951 Run time : 06m 27s CPU time : 00m 29s NOTES -- GUI lag is better then -tt 90 -sbs 256. There remains some minor intermittent keyboard lags for example, but on my rig, it's not a deal breaker. Compared to "stock" (or "BASELINE"), the Run time is ~38% longer and the CPU time is ~33% longer. While testing, I was web surfing, composing this post, and switching between open apps.Presently, testing -use_sleep only. Let me know what other switch combinations you need tested. ID: 59609 ·

AMDave Volunteer tester Send message Joined: 12 Jan 16 Posts: 38 Credit: 289,647 RAC: 0	Message 59610 - Posted: 18 Sep 2016, 18:42:27 UTC >Â Â -tt 30 -sbs 256 -use_sleep Task 24832352 WU true angle range : 0.007806 Run time : 17m 03s CPU time : 04m 17s Task 24832403 WU true angle range : 3.417951 Run time : 06m 25s CPU time : 01m 02s Task 24832178 WU true angle range : 0.010849 Run time : 26m 40s CPU time : 05m 26s Task 24832425 WU true angle range : 3.417951 Run time : 06m 24s CPU time : 00m 15s Task 24831421 WU true angle range : 0.010040 Run time : 26m 15s CPU time : 06m 23s NOTES -- GUI lag is minimal.Â Â There remains some intermittent keyboard lags for example.Â Â Notice the difference in the Run time and the CPU time for ARs 0.007806, 0.010849, & 0.010040.Â Â Overall, this switch combination couldn't best the "BASELINE". ID: 59610 ·

Stephen Volunteer tester Send message Joined: 16 Sep 16 Posts: 20 Credit: 176,425 RAC: 0	Message 59625 - Posted: 19 Sep 2016, 15:46:58 UTC . . To whom it may concern - . . Runtimes: . . Arecibo VLAR: . . . . . . . CUDA42 . . . <=> 86 mins . . . . . . . CUDA50 . . . <=> 85 mins . . . . . . . SoG r3528. . <=> 86 mins . . Arecibo Normal AR: . . . . . . . CUDA42 . . . <=> not available . . . . . . . CUDA50 . . . <=> 35 mins . . . . . . . SoG r3528. . <=> 37 mins . . Arecibo High AR: . . . . . . . CUDA42 . . . <=> 25 mins . . . . . . . CUDA50 . . . <=> 24 mins . . . . . . . SoG r3528. . <=> 25 mins . . Guppies: . . . . . . . CUDA42 . . . <=> 55 mins . . . . . . . CUDA50 . . . <=> 55 mins . . . . . . . SoG r3528. . <=> 56 mins . . These are very much the actual runtimes for each class of WU. Surprisingly on this rig all apps appear to achieve almost identical runtimes though SoG seems to achieve a consistently higher GPU utilisation level than CUDA. There is about a 60 to 90 second increase in some runtimes with SoG but out of 85 mins that is negligible. Mostly there is no lag though when crunching Arecibo VLAR WUs using CUDA50 there is definite lag when moving the cursor about in Boinc Manager, I have not noticed this with SoG but it may be there and simply not detected. . . Computer Core(2) Duo E7600 3GHz with 6GB ram . . GPU is Nvidia GT730 with 2GB GDDR5 ram . . Driver version is 365.19 . . BOINC version 7.6.22 Stephen . ID: 59625 ·

Raistmer Volunteer tester Send message Joined: 18 Aug 05 Posts: 2423 Credit: 15,878,738 RAC: 0	Message 59635 - Posted: 19 Sep 2016, 19:02:46 UTC - in response to Message 59625. . . GPU is Nvidia GT730 with 2GB GDDR5 ram . For such cards BOINC will ultimately select CUDA42 app (if it will measure relative performance correctly). News about SETI opt app releases: https://twitter.com/Raistmer ID: 59635 ·

Raistmer Volunteer tester Send message Joined: 18 Aug 05 Posts: 2423 Credit: 15,878,738 RAC: 0	Message 59647 - Posted: 20 Sep 2016, 8:03:13 UTC - in response to Message 59610. Last modified: 20 Sep 2016, 8:05:06 UTC Â Â Notice the difference in the Run time and the CPU time for ARs 0.007806, 0.010849, & 0.010040.Â That's why the best is to compare exactly same ARs with different switches (or to plot AR-dependence curve for each settings set, that requires much-much more completed tasks and much bigger statistics). "baseline" was WU true angle range : 0.007977 so indirect comparison with latest only possible. News about SETI opt app releases: https://twitter.com/Raistmer ID: 59647 ·

©2023 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.