Message boards :
Number crunching :
Monitoring inconclusive GBT validations and harvesting data for testing
Message board moderation
Previous · 1 . . . 32 · 33 · 34 · 35 · 36 · Next
Author | Message |
---|---|
jason_gee Send message Joined: 24 Nov 06 Posts: 7489 Credit: 91,093,184 RAC: 0 |
HueHueHueHueHueHueHue "Living by the wisdom of computer science doesn't sound so bad after all. And unlike most advice, it's backed up by proofs." -- Algorithms to live by: The computer science of human decisions. |
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
I eagerly await the rollout of the latest SoG and Intel GPU apps to full production status. The bigger beta participation will be the sooner they will appear on main. Eric placed them on beta already: Windows/x86 8.20 (opencl_ati5_sah) 8 Nov 2016, 23:03:25 UTC 0 GigaFLOPS Windows/x86 8.20 (opencl_intel_gpu_sah) 25 Oct 2016, 18:46:05 UTC 175 GigaFLOPS Windows/x86 8.20 (opencl_nvidia_SoG) 8 Nov 2016, 23:03:25 UTC 255 GigaFLOPS BTW, did you test v8.20 SoG vs CPU validation offline for that unfortunate task? Would be interesting to see if 8.20 could help there. SETI apps news We're not gonna fight them. We're gonna transcend them. |
Jeff Buck Send message Joined: 11 Feb 00 Posts: 1441 Credit: 148,764,870 RAC: 0 |
I eagerly await the rollout of the latest SoG and Intel GPU apps to full production status. Okay, I'll switch my multi-GPU boxes over to Beta as soon as this surprise maintenance period is over. That will add 11 NVIDIA GPUs to the mix. Can't help with the ATI and Intel, though. BTW, did you test v8.20 SoG vs CPU validation offline for that unfortunate task? No, I didn't. However, so far I haven't seen a single instance on my machines where r3556 didn't agree with stock CPU, so I feel comfortable that it would have matched up here, as well. If anyone would like to run an offline test with it, though, I've uploaded the WU file to my Amazon cloud drive as WU226235680.zip. |
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
Got it, thanks. Will check with ATi SoGs. SETI apps news We're not gonna fight them. We're gonna transcend them. |
Richard Haselgrove Send message Joined: 4 Jul 99 Posts: 14654 Credit: 200,643,578 RAC: 874 |
Copying from the installer thread: http://setiathome.berkeley.edu/workunit.php?wuid=2320732915 http://boinc2.ssl.berkeley.edu/sah/download_fanout/11d/20ja09ae.12842.7843.8.35.202 Looks like it's the pulses again. (opencl_intel_gpu_sah) x86_64-apple-darwin Best pulse: peak=6.931435 , time=96.44, period=2.586 , d_freq=1419474418.49, score=0.9659, chirp=-25.435, fft_len=512 r3557 SoG Best pulse: peak=0.9932134, time=11.99, period=0.1532, d_freq=1419474804.37, score=0.9999, chirp= 71.006, fft_len=128 Although both are below the reporting threshhold, they're wildly different. |
Jeff Buck Send message Joined: 11 Feb 00 Posts: 1441 Credit: 148,764,870 RAC: 0 |
That v8.00 (opencl_intel_gpu_sah) x86_64-apple-darwin shows up quite a lot in the Inconclusive lists I've been generating for my hosts. For example: Workunit 2320911573 (22mr09ab.23966.2117.9.36.85) Task 5275155431 (S=0, A=1, P=0, T=0, G=0) v8.00 (opencl_intel_gpu_sah) x86_64-apple-darwin Task 5275155432 (S=0, A=1, P=1, T=0, G=0) SSE3xj Win32 Build 3556 The corresponding best pulses are: Best pulse: peak=7.604383, time=12.96, period=2.615, d_freq=1420828933.46, score=0.9786, chirp=29.416, fft_len=64 Best pulse: peak=3.731792, time=25.91, period=1.106, d_freq=1420833535.84, score=1.022, chirp=30.395, fft_len=128 I've generally been ignoring all the Intel GPUs showing up on my list, and now that there's a new version over on Beta, v8.19, I wonder if it's worth paying attention to these here on Main at all. |
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
To save time i would recommend to concentrate on proven valid versus currently in testing and omit proven imprecise. Proven imprecise are imprecise indeed and onyl fix for this - to speedup testing and release of new builds. v8.00 (opencl_intel_gpu_sah) x86_64-apple-darwin is proven imprecise. (that's why we have new TBar's binaries on beta few days already) 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 |
As I already wrote on beta, would be good to put additional CPU processing with proved apps back to beta for some time. To increase number of valid reference results for testing new builds. SETI apps news We're not gonna fight them. We're gonna transcend them. |
tullio Send message Joined: 9 Apr 04 Posts: 8797 Credit: 2,930,782 RAC: 1 |
Both my AMD A10-6700 in Windows 10 and Opteron 1210 in Linux-64 are crunching SETI Beta tasks. Tullio |
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
No, I didn't. However, so far I haven't seen a single instance on my machines where r3556 didn't agree with stock CPU, so I feel comfortable that it would have matched up here, as well. If anyone would like to run an offline test with it, though, I've uploaded the WU file to my Amazon cloud drive as WU226235680.zip. And offline results: MB8_win_x86_SSE2_OpenCL_ATi_HD5_r3330.exe / blc4_2bit_guppi_57449_42556_HIP78775_0009.14621.416.18.27.57.vlar.wu : R2: .\ref\ref-MB8_win_x86_SSE3_VS2008_r3330.exe-blc4_2bit_guppi_57449_42556_HIP78775_0009.14621.416.18.27.57.vlar.wu.res Result : Strongly similar, Q= 99.58% setiathome_8.19_windows_intelx86__opencl_ati5_sah.exe / blc4_2bit_guppi_57449_42556_HIP78775_0009.14621.416.18.27.57.vlar.wu : R2: .\ref\ref-MB8_win_x86_SSE3_VS2008_r3330.exe-blc4_2bit_guppi_57449_42556_HIP78775_0009.14621.416.18.27.57.vlar.wu.res ------------- R1:R2 ------------ ------------- R2:R1 ------------ Exact Super Tight Good Bad Exact Super Tight Good Bad Spike 0 1 1 1 0 0 1 1 1 24 Gaussian 0 0 0 0 0 0 0 0 0 0 Pulse 0 5 5 5 25 0 5 5 5 0 Triplet 0 0 0 0 0 0 0 0 0 0 Best Spike 0 0 0 0 0 0 0 0 0 0 Best Gaussian 0 0 0 0 0 0 0 0 0 0 Best Pulse 0 0 0 0 0 0 0 0 0 0 Best Triplet 0 0 0 0 0 0 0 0 0 0 ---- ---- ---- ---- ---- ---- ---- ---- ---- ---- 0 6 6 6 25 0 6 6 6 24 Unmatched signal(s) in R1 at line(s) 473 500 527 553 579 605 632 659 687 713 740 768 795 822 849 876 903 930 958 985 1011 1037 1065 1091 1118 Unmatched signal(s) in R2 at line(s) 409 425 467 483 499 515 531 547 563 606 649 665 681 697 713 729 745 761 777 793 809 825 841 857 For R1:R2 matched signals only, Q= 99.80% Result : Different. setiathome_8.20_windows_intelx86__opencl_ati5_sah.exe / blc4_2bit_guppi_57449_42556_HIP78775_0009.14621.416.18.27.57.vlar.wu : R2: .\ref\ref-MB8_win_x86_SSE3_VS2008_r3330.exe-blc4_2bit_guppi_57449_42556_HIP78775_0009.14621.416.18.27.57.vlar.wu.res Result : Strongly similar, Q= 99.58% So yes, 8.20 should process such tasks w/o inconclusives. SETI apps news We're not gonna fight them. We're gonna transcend them. |
Jeff Buck Send message Joined: 11 Feb 00 Posts: 1441 Credit: 148,764,870 RAC: 0 |
Way back on 29 Sep, in Message 1820375, I posted the first quadruple Inconclusive that I had run across on one of my machines. Finally, after the original potential tiebreaker host timed out, another host just finished running that WU, and guess what, we have another quintuple Inconclusive. Workunit 2267687414 (blc4_2bit_guppi_57403_HIP11048_0004.3489.0.22.45.87.vlar) Task 5162535378 (S=17, A=0, P=12, T=1, G=0) v8.12 (opencl_nvidia_SoG) windows_intelx86 Task 5162535379 (S=19, A=0, P=10, T=1, G=0) v8.00 windows_intelx86 Task 5164467009 (S=10, A=0, P=19, T=1, G=0) SSE3xj Win32 Build 3500 Task 5167802793 (S=15, A=0, P=14, T=1, G=0) v8.12 (opencl_nvidia_SoG) windows_intelx86 Task 5286353178 (S=6, A=0, P=23, T=1, G=0) v8.19 (opencl_nvidia_SoG) windows_intelx86 Unfortunately, it's now gone out to another host running SoG r3528, so I guess it's a tossup as to whether it will validate or whether the scheduler will finally give up on this WU. :^) |
Jeff Buck Send message Joined: 11 Feb 00 Posts: 1441 Credit: 148,764,870 RAC: 0 |
Just for the record and future reference, the final host did actually validate against the other SoG r3528 with counts of (S=6, A=0, P=23, T=1, G=0). That means that Task 5286353178 was anointed as the canonical result, with counts that are significantly different than the "gold standard" stock Windows CPU app. So it goes. Hopefully v8.20 SoG will be here soon and most of these sorts of messes will go away. |
Jeff Buck Send message Joined: 11 Feb 00 Posts: 1441 Credit: 148,764,870 RAC: 0 |
I think this WU is the first one I've noticed where one of my r3556 tasks got an Inconclusive against what appears to be a stable stock Windows host. The signal counts match, though, so it must be something subtle. Workunit 2332864190 (21fe09ab.7357.18068.5.32.196) Task 5300443647 (S=10, A=0, P=6, T=1, G=1) SSE3xj Win32 Build 3556 Task 5300443648 (S=10, A=0, P=6, T=1, G=1) v8.00 windows_intelx86 |
Oddbjornik Send message Joined: 15 May 99 Posts: 220 Credit: 349,610,548 RAC: 1,728 |
Here's a (possibly) interesting inconclusive between my OpenCL r3557 and Petri33's "x41p_zi3k, Cuda 8.00 special". |
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
Here's a (possibly) interesting inconclusive between my OpenCL r3557 and Petri33's "x41p_zi3k, Cuda 8.00 special". In my counter it's 4th case of same bug demonstration. Pulse: peak=9.964181, time=45.99, period=28.63, d_freq=1544533058.61, score=1.053, chirp=26.651, fft_len=4k Pulse: peak=9.73683, time=45.99, period=27.2, d_freq=1544533058.61, score=1.031, chirp=26.651, fft_len=4k D: threshold 2.258369; unscaled peak power: 2.367302 exceeds threshold for 4.824% As one can see, OpenCL reported well above threshold pulse so - not precision/rounding related case (just as before for these cases). Place of report - the same (time/freq). But periods (and powers, but in this case powers just consequence of periods) are different. Absolutely same case as 3 before. Wrong pulse picked up between all that found on different folding patterns in that place. I suggested place to look in very first occurence of this issue. So, 4th confirmation of bug existence here. 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 |
I think this WU is the first one I've noticed where one of my r3556 tasks got an Inconclusive against what appears to be a stable stock Windows host. The signal counts match, though, so it must be something subtle. Validated by third SoG. Spike: peak=24.80969, time=87.24, d_freq=1419411785.12, chirp=1.4215, fft_len=128k Spike: peak=24.80969, time=87.24, d_freq=1419411785.12, chirp=1.4215, fft_len=128k Spike: peak=25.3747, time=87.24, d_freq=1419411785.12, chirp=1.4224, fft_len=128k Spike: peak=25.3747, time=87.24, d_freq=1419411785.12, chirp=1.4224, fft_len=128k Spike: peak=24.8094, time=87.24, d_freq=1419411785.13, chirp=1.4234, fft_len=128k Spike: peak=24.8094, time=87.24, d_freq=1419411785.13, chirp=1.4234, fft_len=128k Spike: peak=24.04475, time=87.24, d_freq=1419415385.76, chirp=5.9282, fft_len=128k Spike: peak=24.04474, time=87.24, d_freq=1419415385.76, chirp=5.9282, fft_len=128k Spike: peak=25.90574, time=87.24, d_freq=1419415385.77, chirp=5.9291, fft_len=128k Spike: peak=25.90574, time=87.24, d_freq=1419415385.77, chirp=5.9291, fft_len=128k Spike: peak=26.56526, time=87.24, d_freq=1419415385.78, chirp=5.9301, fft_len=128k Spike: peak=26.56526, time=87.24, d_freq=1419415385.78, chirp=5.9301, fft_len=128k Spike: peak=25.94405, time=87.24, d_freq=1419415385.78, chirp=5.931, fft_len=128k Spike: peak=25.94406, time=87.24, d_freq=1419415385.78, chirp=5.931, fft_len=128k Spike: peak=24.13967, time=87.24, d_freq=1419415385.79, chirp=5.9319, fft_len=128k Spike: peak=24.13967, time=87.24, d_freq=1419415385.79, chirp=5.9319, fft_len=128k Pulse: peak=2.539017, time=5.99, period=0.6969, d_freq=1419416468.8, score=1.012, chirp=19.613, fft_len=256 Pulse: peak=2.539017, time=5.99, period=0.6969, d_freq=1419416468.8, score=1.012, chirp=19.613, fft_len=256 D: threshold 0.0539219; unscaled peak power: 0.05440334 exceeds threshold for 0.8928% Pulse: peak=2.53191, time=5.99, period=0.6969, d_freq=1419416471.97, score=1.01, chirp=20.142, fft_len=256 Pulse: peak=2.531911, time=5.99, period=0.6969, d_freq=1419416471.97, score=1.01, chirp=20.142, fft_len=256 D: threshold 0.0540199; unscaled peak power: 0.05439276 exceeds threshold for 0.6902% Spike: peak=24.02079, time=46.98, d_freq=1419409835.64, chirp=21.01, fft_len=128k Spike: peak=24.02079, time=46.98, d_freq=1419409835.64, chirp=21.01, fft_len=128k Spike: peak=24.58228, time=46.98, d_freq=1419409835.65, chirp=21.012, fft_len=128k Spike: peak=24.58227, time=46.98, d_freq=1419409835.65, chirp=21.012, fft_len=128k Triplet: peak=9.369192, time=6.016, period=4.981, d_freq=1419416738.68, chirp=-24.383, fft_len=256 Triplet: peak=9.369189, time=6.016, period=4.981, d_freq=1419416738.68, chirp=-24.383, fft_len=256 Gaussian: peak=3.675008, mean=0.5456764, ChiSq=1.406728, time=78.01, d_freq=1419418787.66, Gaussian: peak=3.675009, mean=0.5456763, ChiSq=1.406729, time=78.01, d_freq=1419418787.66, score=0.4186165, null_hyp=2.268583, chirp=69.622, fft_len=16k score=0.418625, null_hyp=2.268584, chirp=69.622, fft_len=16k Pulse: peak=6.248314, time=11.97, period=2.235, d_freq=1419416334.2, score=1.017, chirp=-71.56, fft_len=256 Pulse: peak=6.248315, time=11.97, period=2.235, d_freq=1419416334.2, score=1.017, chirp=-71.56, fft_len=256 D: threshold 0.1123426; unscaled peak power: 0.1139427 exceeds threshold for 1.424% Pulse: peak=3.591187, time=17.94, period=1.157, d_freq=1419416042.6, score=1.008, chirp=74.21, fft_len=256 Pulse: peak=3.591187, time=17.94, period=1.157, d_freq=1419416042.6, score=1.008, chirp=74.21, fft_len=256 D: threshold 0.07106485; unscaled peak power: 0.07150516 exceeds threshold for 0.6196% Pulse: peak=3.653561, time=17.94, period=1.157, d_freq=1419416052.02, score=1.025, chirp=76.861, fft_len=256 Pulse: peak=3.653561, time=17.94, period=1.157, d_freq=1419416052.02, score=1.025, chirp=76.861, fft_len=256 D: threshold 0.06976143; unscaled peak power: 0.07114729 exceeds threshold for 1.987% Pulse: peak=6.321055, time=17.94, period=2.313, d_freq=1419416061.42, score=1.027, chirp=79.51, fft_len=256 Pulse: peak=6.321057, time=17.94, period=2.313, d_freq=1419416061.42, score=1.027, chirp=79.51, fft_len=256 D: threshold 0.1113693; unscaled peak power: 0.113965 exceeds threshold for 2.331% Best spike: peak=26.56526, time=87.24, d_freq=1419415385.78, chirp=5.9301, fft_len=128k Best spike: peak=26.56526, time=87.24, d_freq=1419415385.78, chirp=5.9301, fft_len=128k Best autocorr: peak=17.27914, time=87.24, delay=3.0895, d_freq=1419411483.18, chirp=-29.565, fft_len=128k Best autocorr: peak=17.27913, time=87.24, delay=3.0895, d_freq=1419411483.18, chirp=-29.565, fft_len=128k Best gaussian: peak=3.675008, mean=0.5456764, ChiSq=1.406728, time=78.01, d_freq=1419418787.66, Best gaussian: peak=3.675009, mean=0.5456763, ChiSq=1.406729, time=78.01, d_freq=1419418787.66, score=0.4186165, null_hyp=2.268583, chirp=69.622, fft_len=16k score=0.418625, null_hyp=2.268584, chirp=69.622, fft_len=16k Best pulse: peak=6.321056, time=17.94, period=2.313, d_freq=1419416061.42, score=1.027, chirp=79.51, fft_len=256 Best pulse: peak=6.321057, time=17.94, period=2.313, d_freq=1419416061.42, score=1.027, chirp=79.51, fft_len=256 Best triplet: peak=9.369192, time=6.016, period=4.981, d_freq=1419416738.68, chirp=-24.383, fft_len=256 Best triplet: peak=9.369189, time=6.016, period=4.981, d_freq=1419416738.68, chirp=-24.383, fft_len=256 After reordering both SoG looks identical. Hard to say what CPU stock reported. SETI apps news We're not gonna fight them. We're gonna transcend them. |
Jeff Buck Send message Joined: 11 Feb 00 Posts: 1441 Credit: 148,764,870 RAC: 0 |
After reordering both SoG looks identical. Hard to say what CPU stock reported. Okay, I just uploaded the Workunit file to a cloud drive, if you want to try an offline test. |
djmotiska Send message Joined: 26 Jul 01 Posts: 20 Credit: 29,378,647 RAC: 105 |
Raistmer, do you or some other dev happen to need some of these workunits? I'm running rather old r3330 on ATI and SSE3xj r3330 on CPU which have proven to be very stable. I can save the workunit files and results created by my PC if needed. blc2_2bit_guppi_57424_82409_HIP9598_OFF_0010.2706.831.18.27.9.vlar blc2_2bit_guppi_57424_84052_HIP9727_0015.16679.831.18.27.24.vlar blc3_2bit_guppi_57424_41210_HIP63608_0013.29546.416.18.27.238.vlar blc3_2bit_guppi_57451_27376_HIP69732_OFF_0026.23989.831.17.26.57.vlar blc6_2bit_guppi_57398_MESSIER031_0017.17065.0.23.46.202 blc6_2bit_guppi_57397_MESSIER031_0011.9745.0.23.46.153 02ja09ab.27298.85144.11.38.219 09ja09ab.2683.11524.15.42.26 12mr09aa.6314.22158.4.31.166 20mr09aa.28833.7843.6.33.135 Then there is this one where two ATI apps disagreed on the outcome: 30dc09aj.2161.11528.4.31.30 Edit: got two more resends while I was composing my message: blc4_2bit_guppi_57424_39246_HIP63510_0007.28578.0.18.27.88.vlar blc4_2bit_guppi_57424_40226_HIP63608_OFF_0010.24350.0.17.26.145.vlar |
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
Unfortunately I have no time to look at every inconclusive by myself. Most of them either from too overclocked hardware or from older apps with alerady known limitations in validation behavior. So would be good to filter those inconclusives by "interesting" score. For older revs that run on main - it's definitely non-overflowed result with strong enough signals (to exclude some rounding noise) paired with known-to-be-stable host. Such results if they disagree between each other could represent some source for further bug-hunting. SETI apps news We're not gonna fight them. We're gonna transcend them. |
Jeff Buck Send message Joined: 11 Feb 00 Posts: 1441 Credit: 148,764,870 RAC: 0 |
I'm going to go ahead and post this one since it's now the second recent example of r3556 coming up Inconclusive against an apparently stable host running the stock Windows CPU app. Again, the signal counts appear to match so it's not clear what's causing the Inconclusive. Workunit 2333076601 (21fe09ab.20049.19704.7.34.20) Task 5300892747 (S=0, A=3, P=1, T=3, G=1) v8.00 windows_intelx86 Task 5300892748 (S=0, A=3, P=1, T=3, G=1) SSE3xj Win32 Build 3556 Note that the WU was split from the same tape, 21fe09ab, as the earlier example I posted a couple days ago. Don't know if that might be significant or not. |
©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.