Message boards :
Number crunching :
Inconclusive Work Units Running AP Ver 6
Message board moderation
Previous · 1 . . . 6 · 7 · 8 · 9 · 10 · Next
Author | Message |
---|---|
Josef W. Segur Send message Joined: 30 Oct 99 Posts: 4504 Credit: 1,414,761 RAC: 0 |
yesJust got one, my computer reported minutes ago. I did download and run an offline check on that, just one of the 30 repetitive pulses was questionable, so a weakly similar credit grant was certain. The Lunatics FFA code is vectorized while stock isn't, so there will always be some very small differences leading to occasional disagreement. Joe |
JohnDK Send message Joined: 28 May 00 Posts: 1222 Credit: 451,243,443 RAC: 1,127 |
|
Urs Echternacht Send message Joined: 15 May 99 Posts: 692 Credit: 135,197,781 RAC: 211 |
I have one, Linux stock vs r1305 Your link points at a wu done by two stock OSX and one NV opencl wingcrew. Is the link correct ? _\|/_ U r s |
JohnDK Send message Joined: 28 May 00 Posts: 1222 Credit: 451,243,443 RAC: 1,127 |
I have one, Linux stock vs r1305 Yes, thought it was Linux but it may be OSX. |
Stick Send message Joined: 26 Feb 00 Posts: 100 Credit: 5,283,449 RAC: 5 |
Another one: wuid=999173464 |
Josef W. Segur Send message Joined: 30 Oct 99 Posts: 4504 Credit: 1,414,761 RAC: 0 |
Another one: wuid=999173464 That's fairly unusual: single pulses: 30 repetitive pulses: 1 I'll run an offline test on it, most likely there really are many single pulses but even CPU processing can sometimes glitch and manufacture false pulses. Joe |
Stick Send message Joined: 26 Feb 00 Posts: 100 Credit: 5,283,449 RAC: 5 |
Another one: wuid=999173464 The 3rd task has reported and, surprisingly, all three validated/received credit. |
Josef W. Segur Send message Joined: 30 Oct 99 Posts: 4504 Credit: 1,414,761 RAC: 0 |
Another one: wuid=999173464 Not really surprising, the more signals there are the more likely for a very small difference in calculations to make a difference. My offline run with ap_6.01r557_SSE2_331_AVX.exe finished and found the 30 single and 1 repetitive pulses, the run with stock 6.01 is at about 57% progress and has found 14 single and 1 repetitive all of which match very closely. It should finish about when I get up tomorrow, then I'll check in detail. Joe |
Sirius B Send message Joined: 26 Dec 00 Posts: 24879 Credit: 3,081,182 RAC: 7 |
scrub this..looked at wrong list.. apologies. |
Josef W. Segur Send message Joined: 30 Oct 99 Posts: 4504 Credit: 1,414,761 RAC: 0 |
Another one: wuid=999173464 As it turned out, the stock app's 26th single pulse was only 0.0036% above threshold, and r557 must have calculated it as just below threshold so didn't report it. That wouldn't directly affect validation because only single pulses which are at least 1% above threshold are considered, but indirectly it meant that the 30th single pulse from stock was only the 29th for r557. So r557 actually analyzed that pulse better, stock quit when it first got above threshold by 0.27% but r557 found it was stronger at a higher dispersion and reported it as about 3% above threshold. The Validator calls that a mismatch since it only considers the r557 version. There was another mismatch because r557's 30th single pulse was 3.25% above threshold so also included in the Validator's checking, so two of the pulses checked from the r557 result weren't found in the stock result. Still, in terms of how the Validator works, 27 of 29 checked signals did match so the r557 result was granted credit as being "weakly similar". Joe |
Stick Send message Joined: 26 Feb 00 Posts: 100 Credit: 5,283,449 RAC: 5 |
Joe, Thank you for the explanation! Stick As it turned out, the stock app's 26th single pulse was only 0.0036% above threshold, and r557 must have calculated it as just below threshold so didn't report it. That wouldn't directly affect validation because only single pulses which are at least 1% above threshold are considered, but indirectly it meant that the 30th single pulse from stock was only the 29th for r557. So r557 actually analyzed that pulse better, stock quit when it first got above threshold by 0.27% but r557 found it was stronger at a higher dispersion and reported it as about 3% above threshold. The Validator calls that a mismatch since it only considers the r557 version. There was another mismatch because r557's 30th single pulse was 3.25% above threshold so also included in the Validator's checking, so two of the pulses checked from the r557 result weren't found in the stock result. Still, in terms of how the Validator works, 27 of 29 checked signals did match so the r557 result was granted credit as being "weakly similar".Joe |
Fred E. Send message Joined: 22 Jul 99 Posts: 768 Credit: 24,140,697 RAC: 0 |
|
Josef W. Segur Send message Joined: 30 Oct 99 Posts: 4504 Credit: 1,414,761 RAC: 0 |
Not sure if you're still looking for these, but I have one: If it were practical, we'd analyze all cases of inconclusive results from Lunatics apps. There's too little time and/or equipment for that, but I'll check this one too on general principles. Should be done sometime Sunday. The WU may well be resoved before then and I expect as usual all completed tasks will get credit, but there's always the possibility analysis will point to some lurking bug. Joe |
perryjay Send message Joined: 20 Aug 02 Posts: 3377 Credit: 20,676,751 RAC: 0 |
Here's one from me too... http://setiathome.berkeley.edu/workunit.php?wuid=997551999 PROUD MEMBER OF Team Starfire World BOINC |
Josef W. Segur Send message Joined: 30 Oct 99 Posts: 4504 Credit: 1,414,761 RAC: 0 |
Not sure if you're still looking for these, but I have one: That one turned out to be simple to analyze, the stock app found one more repetitive pulse than optimized. The peak power was 5170.0068359375 and the threshold 5170 (FFA thresholds are truncated to integer values). So it's just one of the inevitable cases where a tiny difference in calculation is enough to cause an inconclusive validation. Joe |
fataldog187 Send message Joined: 4 Nov 02 Posts: 42 Credit: 1,271,261 RAC: 0 |
I couldn't read the whole thread but... I have 3 APv6 with 30 repetitive pulses that are inconclusive http://setiathome.berkeley.edu/workunit.php?wuid=997100145 http://setiathome.berkeley.edu/workunit.php?wuid=1008642492 http://setiathome.berkeley.edu/workunit.php?wuid=1007059107 And one was marked invalid after being sent to a 3rd wingman http://setiathome.berkeley.edu/workunit.php?wuid=1005838187 There are also two others with 30 repetitive pulses that are awaiting validation, and these are all out of only 8 that are listed with AstroPulse v6 Anonymous platform, something seems a little glitchy |
Josef W. Segur Send message Joined: 30 Oct 99 Posts: 4504 Credit: 1,414,761 RAC: 0 |
I couldn't read the whole thread but... Although the r555 ATi OpenCL application has been reliable on most systems, there are exceptions. You might be able to stabilize processing by tuning voltages, clock rates, or command line parameters. Joe |
Fred E. Send message Joined: 22 Jul 99 Posts: 768 Credit: 24,140,697 RAC: 0 |
That one turned out to be simple to analyze, the stock app found one more repetitive pulse than optimized. The peak power was 5170.0068359375 and the threshold 5170 (FFA thresholds are truncated to integer values). So it's just one of the inevitable cases where a tiny difference in calculation is enough to cause an inconclusive validation. Thanks for the analysis - thought it might be something like that since I ususally don't lose when results go inconclusive. I probably won't have any more for a while. I've stopped fetching AP6 until I can figure out what's causing errors with zero runtime - have hit 4 in the past few days and I'm in troubleshoot mode. May open a thread if I don't find a problem Another Fred Support SETI@home when you search the Web with GoodSearch or shop online with GoodShop. |
Wedge009 Send message Joined: 3 Apr 99 Posts: 451 Credit: 431,396,357 RAC: 553 |
Over the past couple of months, I've also been trying to make sense out of the increasing occurrences of invalid results due to returning 30 repetitive pulses. In fact, aside from my main host (running AP/OpenCL on a HD 6950), I think all of my GPU-related AP work-units are returning 30 repetitive pulse results, which I find quite worrying. Especially since I generally don't over-clock and am pretty sure the hosts are otherwise stable. It seems very unlikely to me that so many hosts would start failing simultaneously. I tried doing a full shut-down, rest, and cold boot on some of the affected hosts, as suggested elsewhere for flushing the GPU RAM - there doesn't seem to be new AP work being generated at the moment, so will have to wait to see if that makes any difference. Host 1504137 Main machine, currently only host consistently returning valid AP results. Radeon HD 6950 running OpenCL r555, ironically the only AP-related GPU being over-clocked (albeit very modestly). Host 6077487 A machine I made for my brother, was previously running Brook r555 on a Radeon HD 4850 at factory clock rates. Seemed stable in APv5 era, but have now switched to AP SSE2 r557. Has been returning valid AP work-units since. Host 4462068 A 2008 laptop, was previously running Brook r555 on a Mobility Radeon HD 2600. Again, was fine in APv5 era, but even if it hadn't started generating lots of 30 repetitive pulse results, I would have switched to AP SSE2 r557 anyway because the heat pipe cooling the GPU no longer seems effective (CPU is cooled, but GPU overheats with anything more strenuous than processing a video and shuts everything down). Host 6649080 A fairly new host, albeit older hardware, running OpenCL r555 on HD 4670. Was returning good results initially, but now I notice it is returning mostly 30 repetitive pulses, which mostly seem to be marked as invalid. Host 6692553 Another new host running OpenCL r555 on a Fusion C-50. Before anyone questions whether it's worth the bother on such a low-power machine, I think 20 hours is a big improvement on 100+ hours on an Atom. Anyway, only the first AP work-unit was valid, subsequent results seem to exhibit 30 repetitive pulses. Has also been unlucky with regards to CPU time - a lot of high blanking percentage work-units there. So, basically, I'm not sure what's going on, so was trying to search the forums for any clue on why I'm getting so many 30 repetitive pulse results. Soli Deo Gloria |
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
There was bugfix in r1305 that potentially can improve app precision. All who participate please upgrade to "stock" 6.03 binary for AP (it's rev1305 binary). All who dont' run beta - await for binary update... |
©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.