Message boards :
Number crunching :
Linux CUDA 'Special' App finally available, featuring Low CPU use
Message board moderation
Previous · 1 . . . 40 · 41 · 42 · 43 · 44 · 45 · 46 . . . 83 · Next
Author | Message |
---|---|
Brent Norman Send message Joined: 1 Dec 99 Posts: 2786 Credit: 685,657,289 RAC: 835 |
Ahh yes, I removed -pfb 64 when you mentioned that, but I haven't run many since then. I did run 50 through without it during maintenance though. |
jason_gee Send message Joined: 24 Nov 06 Posts: 7489 Credit: 91,093,184 RAC: 0 |
Not quite yet, though I have a copy. Am using this weekend to do that, now that switchover to faster internet is done and teething problems ironed out. [Edit:] Well that goes much quicker with 20Mbps upload, instead of 1Mbps :) Cuda multibeam Alpha, Updated client/alpha to Petri's zi3v. Addresses the pulse race condition. Some questions about best Gaussian are being investigated. "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. |
TBar Send message Joined: 22 May 99 Posts: 5204 Credit: 840,779,836 RAC: 2,768 |
Ahh yes, I removed -pfb 64 when you mentioned that, but I haven't run many since then.Looks like there's another one from a few days ago, https://setiweb.ssl.berkeley.edu/beta/workunit.php?wuid=9835221 I downloaded the task here, http://boinc2.ssl.berkeley.edu/beta/download/3ec/04oc08ab.11260.17659.13.47.23 It's going the be a couple hours before the CPU is finished with it. I was able to create a Working Win To Go Portable USB drive. Seems to work as advertised including the computer's internal drive being ignored. The problem is I couldn't get the clone function to work so the install ended up being a normal new install needing everything reinstalled, including all those freakin' updates I just got finished installing on the normal install. I think I'll just swap drives between machines and put the normal install on the ATI machine. Then use the portable on the other machine if I need to run Windows for anything. So, I guess it is possible to install Windows 8.1 on two different machines, as long as 1 is installed as a Win To Go USB drive. |
TBar Send message Joined: 22 May 99 Posts: 5204 Credit: 840,779,836 RAC: 2,768 |
So, My CPU showed the 1080 as correct on the task, https://setiweb.ssl.berkeley.edu/beta/workunit.php?wuid=9835221 WU true angle range is : 0.446565 Spike: peak=24.35858, time=6.711, d_freq=1420223558.53, chirp=-12.258, fft_len=128k Spike: peak=24.25936, time=73.82, d_freq=1420220528.11, chirp=20.491, fft_len=128k Autocorr: peak=18.02235, time=60.4, delay=3.1659, d_freq=1420226224.46, chirp=26.741, fft_len=128k Triplet: peak=9.61119, time=42.6, period=2.825, d_freq=1420219875.63, chirp=-26.948, fft_len=128 Pulse: peak=5.248671, time=23.93, period=1.665, d_freq=1420227266.41, score=1.02, chirp=-44.385, fft_len=512 Pulse: peak=0.938087, time=48.09, period=0.1335, d_freq=1420220338.92, score=1.02, chirp=-50.726, fft_len=16 Pulse: peak=0.9314071, time=48.09, period=0.1335, d_freq=1420220339.95, score=1.013, chirp=-76.089, fft_len=16 Pulse: peak=7.605646, time=84.05, period=2.63, d_freq=1420220017.16, score=1.006, chirp=82.429, fft_len=128 Best spike: peak=24.35858, time=6.711, d_freq=1420223558.53, chirp=-12.258, fft_len=128k Best autocorr: peak=18.02235, time=60.4, delay=3.1659, d_freq=1420226224.46, chirp=26.741, fft_len=128k Best gaussian: peak=3.495132, mean=0.5419289, ChiSq=1.328668, time=69.63, d_freq=1420220651.02, score=-0.6907406, null_hyp=2.161481, chirp=83.305, fft_len=16k Best pulse: peak=0.938087, time=48.09, period=0.1335, d_freq=1420220338.92, score=1.02, chirp=-50.726, fft_len=16 Best triplet: peak=9.61119, time=42.6, period=2.825, d_freq=1420219875.63, chirp=-26.948, fft_len=128 Spike count: 2 Autocorr count: 1 Pulse count: 4 Triplet count: 1 Gaussian count: 0 Time cpu in use since last restart: 8892.5 seconds Don't know, I found one of mine one count high against a CPU, https://setiathome.berkeley.edu/workunit.php?wuid=2588520990 My CPU says; Spike count: 0 Autocorr count: 0 Pulse count: 10 Triplet count: 0 Gaussian count: 0 Perhaps there is still a little strangeness with the PulseFind...but it's rare. |
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
Don't know, I found one of mine one count high against a CPU, https://setiathome.berkeley.edu/workunit.php?wuid=2588520990 And task re-run offline on GPU? SETI apps news We're not gonna fight them. We're gonna transcend them. |
TBar Send message Joined: 22 May 99 Posts: 5204 Credit: 840,779,836 RAC: 2,768 |
I decided to run some of the recent failed tasks again on the Mac using 2 different Apps. The CUDA75 zi3v version doesn't like the Unroll fix when using the 7.5 driver in Yosemite, but seems to work better in Sierra with the Cuda 8.0 driver than the CUDA80 App...weird. Most of the tasks are running fine except there is definitely something going on with the Triplets in this task, https://setiweb.ssl.berkeley.edu/beta/workunit.php?wuid=9837856. The Mac Apps didn't like it any better than the Linux App on the 750Ti. Another task, 04oc08ab.15453.21340.13.47.244, gave One bad Triplet. The task 04oc08ab.31484.2526.13.47.116 is another one giving a Bad Best Pulse. The task 23se08ac.6875.22968.6.33.135 worked again on the Mac with zi3v even though it Fails in Linux. Lastly, blc05_2bit_guppi_57834_84204_HIP26017_0003.15096.818.24.47.37.vlar was One Pulse too high with both Apps. Current WU: blc05_2bit_guppi_57834_84204_HIP26017_0003.15096.818.24.47.37.vlar.wu --------------------------------------------------- Skipping default app MBv8_8.05r3344_sse41_x86_64-apple-darwin, displaying saved result(s) Elapsed Time: ………………………………… 7498 seconds --------------------------------------------------- Running app with command : setiathome_x41p_zi3v_x86_64-apple-darwin_cuda75 -device 0 556.31 real 66.76 user 24.59 sys Elapsed Time : ……………………………… 556 seconds Speed compared to default : 1348 % ----------------- Comparing results ------------- R1:R2 ------------ ------------- R2:R1 ------------ Exact Super Tight Good Bad Exact Super Tight Good Bad Spike 0 0 0 0 0 0 0 0 0 0 Autocorr 0 0 0 0 0 0 0 0 0 0 Gaussian 0 0 0 0 0 0 0 0 0 0 Pulse 0 10 10 10 0 0 10 10 10 1 Triplet 0 0 0 0 0 0 0 0 0 0 Best Spike 0 1 1 1 0 0 1 1 1 0 Best Autocorr 0 1 1 1 0 0 1 1 1 0 Best Gaussian 1 1 1 1 0 1 1 1 1 0 Best Pulse 0 1 1 1 0 0 1 1 1 0 Best Triplet 0 0 0 0 0 0 0 0 0 0 ---- ---- ---- ---- ---- ---- ---- ---- ---- ---- 1 14 14 14 0 1 14 14 14 1 Unmatched signal(s) in R2 at line(s) 450 For R1:R2 matched signals only, Q= 99.23% Result : Weakly similar. --------------------------------------------------- Running app with command : setiathome_x41p_zi3v_x86_64-apple-darwin_cuda80 -device 0 556.54 real 66.99 user 24.98 sys Elapsed Time : ……………………………… 557 seconds Speed compared to default : 1346 % ----------------- Comparing results ------------- R1:R2 ------------ ------------- R2:R1 ------------ Exact Super Tight Good Bad Exact Super Tight Good Bad Spike 0 0 0 0 0 0 0 0 0 0 Autocorr 0 0 0 0 0 0 0 0 0 0 Gaussian 0 0 0 0 0 0 0 0 0 0 Pulse 0 10 10 10 0 0 10 10 10 1 Triplet 0 0 0 0 0 0 0 0 0 0 Best Spike 0 1 1 1 0 0 1 1 1 0 Best Autocorr 0 1 1 1 0 0 1 1 1 0 Best Gaussian 1 1 1 1 0 1 1 1 1 0 Best Pulse 0 1 1 1 0 0 1 1 1 0 Best Triplet 0 0 0 0 0 0 0 0 0 0 ---- ---- ---- ---- ---- ---- ---- ---- ---- ---- 1 14 14 14 0 1 14 14 14 1 Unmatched signal(s) in R2 at line(s) 450 For R1:R2 matched signals only, Q= 99.23% Result : Weakly similar. --------------------------------------------------- Done with blc05_2bit_guppi_57834_84204_HIP26017_0003.15096.818.24.47.37.vlar.wu. In other Current Events, it appears that the Crunchers Anonymous move to the new Host has Corrupted ALL the Existing Downloads. Not only is the Maximum File size now only 2.5 MBs instead of 20, but, All the Existing files are going to have to be replaced. Hopefully they can be updated soon, some already are. |
Brent Norman Send message Joined: 1 Dec 99 Posts: 2786 Credit: 685,657,289 RAC: 835 |
I see another mismatch for pulses https://setiweb.ssl.berkeley.edu/beta/workunit.php?wuid=9855192 I will ran another 50 through with -unroll autotune -nobs ... |
TBar Send message Joined: 22 May 99 Posts: 5204 Credit: 840,779,836 RAC: 2,768 |
That's an overflow, SETI@Home Informational message -9 result_overflow They don't really count as it just means there are more signals than counted and one machine found different signals before they were stopped. If all the signals were counted, it's possible they would be the same. It seems Summer has hit Finland. When's the last time you didn't see Petri's machine #1? https://setiathome.berkeley.edu/top_hosts.php Oh, how about those Raccoons? https://www.seti-germany.de/Wow/stats_team_all.php?year=&lang=en BTW, if anyone can build a working Copy of BOINC 7.5+ that works in the Home Folder of Ubuntu 16.04 Please submit it to Berkeley. I tried it last year and gave up after a few weeks. |
Zalster Send message Joined: 27 May 99 Posts: 5517 Credit: 528,817,460 RAC: 242 |
Oh, how about those Raccoons? Still 43 days before that thing starts, lol... And summer has definitely arrived in the Northern Hemisphere.... |
TBar Send message Joined: 22 May 99 Posts: 5204 Credit: 840,779,836 RAC: 2,768 |
Yes, still plenty of time to get a machine up and running with Petri's Special App. After some tweaking I was able to upload files up to 7 MBs at C.A. So, the new CUDA 8.0 download has all the support Apps in it again, Linux_zi3v-CUDA80_Special.7z The zi3v source code is now at the SETI Repository, https://setisvn.ssl.berkeley.edu/trac/browser/branches/sah_v7_opt/Xbranch/client/alpha/PetriR_raw3 |
Brent Norman Send message Joined: 1 Dec 99 Posts: 2786 Credit: 685,657,289 RAC: 835 |
Thanks TBarr, I missed that it was an overflow so no problems there. |
Stephen "Heretic" Send message Joined: 20 Sep 12 Posts: 5557 Credit: 192,787,363 RAC: 628 |
Yes, still plenty of time to get a machine up and running with Petri's Special App. . . Hi TBar, . . I installed this new one with the defaults but I have no idea of how it performs on the 970s because these damned SETI servers will not send me work ... :( In an hour and a half all I managed tot get from it was 2 AP tasks which seemed to be similar to the Windows times (but then it is basically the same ap, OpenCL 100). I will try again later when I get time. Stephen :( |
Stephen "Heretic" Send message Joined: 20 Sep 12 Posts: 5557 Credit: 192,787,363 RAC: 628 |
. . Hi again, . . OK, this time I am getting tasks from the servers (phew). On Arecibo (norm) tasks runtimes are very consistent at 3.2 to 3.3 mins on the 970s. For the more common (ATM) VHAR tasks runtimes are all within 1 or 2 secs of 1.5 mins. So they are very similar to (as I had expected) the runtimes I am getting on the 1060s (even though that machine is driven by a clunky old Pentium-D and this one by an i5-6600). I am not crunching on any CPU cores for this test and the CPU usage is varying between 20% and about 36-38%, so it is just that bit too high to consider crunching on 3 CPU cores, I would only use 2 to allow for some headroom. . . After some time I will change from the defaults to turn off BS and see what difference that makes to run times and CPU usage and let you know. . . Thanks for packaging the AP and MB-CPU aps with the new release. Though I will have to find my way to Arkayn's treasure trove and find an AVX ap for the long term. Stephen :) |
Juha Send message Joined: 7 Mar 04 Posts: 388 Credit: 1,857,738 RAC: 0 |
BTW, if anyone can build a working Copy of BOINC 7.5+ that works in the Home Folder of Ubuntu 16.04 Please submit it to Berkeley. I tried it last year and gave up after a few weeks. It looks like we are getting serious about releasing 7.8, or at least testing it. I suppose I could build it once it gets nearer to recommended status. |
Stephen "Heretic" Send message Joined: 20 Sep 12 Posts: 5557 Credit: 192,787,363 RAC: 628 |
. . @ TBar . . OK, I have been running with BS off for about 16 hours and the results blow my tiny little mind. . . GPU power consumption up about 10 to 15%, productivity up about 15%. GPU temp on GPU-0 now into the 60 C range even with fans at 90% in winter, keeping the room warm :). CPU use holding steady at 52%, I don't think I will bother crunching on the spare CPU core because normal AR Arecibo tasks now run in under 2 mins consistently (like clockwork). The app is finished in under 110 secs but BOINC/manager takes another 6 to 10 secs to close files and move the results to the upload Q. My humble little machine is churning out about 60 plus results per hour . . . Hoooeeeyyyy Bob! 8^} . . The flies in the ointment are Guppis. Blc05 tasks are still taking about 5.2 to 5.4 mins :( Oh well nothing is perfect. :) . . The 970s are doing significantly better than 1060s when the stops are pulled out. Now I will wonder how much improvement the 1060s might get from a Ryzen uograde ... 8^} . . Stephen Yahooooeeey! |
petri33 Send message Joined: 6 Jun 02 Posts: 1668 Credit: 623,086,772 RAC: 156 |
Hi Stephen, You could try to Ryzenize your other CPUs just by testing with no CPU tasks and no blocking sync for GPU tasks. EDIT: and I looked at your computers and saw that you have done that. Petri To overcome Heisenbergs: "You can't always get what you want / but if you try sometimes you just might find / you get what you need." -- Rolling Stones |
TBar Send message Joined: 22 May 99 Posts: 5204 Credit: 840,779,836 RAC: 2,768 |
It would be greatly appreciated. I just tried it again with both boinc-master 7.7 and 7.5. It's a nightmare. The compiler can't find any of the files that are in folders with numbers, such as glib-2.0 or gtx-2.0, without making links to the contained folders. Nothing I do will convince the compiler I actually have libnotify and gtk+-2.0pc right where they're supposed to be, and apparently all the .cpp files in clientgui are missing a {; ...according to the compiler. After that I hit another snag and gave up...again. From the last time I think I remember someone saying there was a copy of wx-3.0 somewhere that worked a little better. But, I can't seem to find it now. Surely there is an easier way.BTW, if anyone can build a working Copy of BOINC 7.5+ that works in the Home Folder of Ubuntu 16.04 Please submit it to Berkeley. I tried it last year and gave up after a few weeks. :-< |
TBar Send message Joined: 22 May 99 Posts: 5204 Credit: 840,779,836 RAC: 2,768 |
I found the version of wx3 I was looking for here, https://launchpad.net/~costamagnagianfranco/+archive/ubuntu/locutusofborg-ppa/+build/9539157 Tried it with a new download of bonic-master and ended up at the same place I stopped the last time. I started a new thread if someone has anything to offer, https://boinc.berkeley.edu/dev/forum_thread.php?id=11719 Doesn't look as though there's much activity over there. |
MarkJ Send message Joined: 17 Feb 08 Posts: 1139 Credit: 80,854,192 RAC: 5 |
I found the version of wx3 I was looking for here, https://launchpad.net/~costamagnagianfranco/+archive/ubuntu/locutusofborg-ppa/+build/9539157 Not sure if you have tried dropping Gianfranco an email. He is one of the Debian maintainers. I know when DA announced a test version on BOINC-alpha Gianfranco asked for things to get fixed up. I think the whole dev tree is stuffed at the moment. BOINC blog |
Stephen "Heretic" Send message Joined: 20 Sep 12 Posts: 5557 Credit: 192,787,363 RAC: 628 |
. . @ TBar . . Oh well I was wrong yet again, the tasks that were completing in about 2 mins were not quite normal Arecibo tasks, their ARs were rather higher than average but not in the VHAR realm. . . Having been able to look at the STDERR files for the results on some of them the numbers are more like this ... VHAR AR > 1.1 1.3 mins QHAR AR ~ 0.7 to 0.9 1.9 mins NARA AR ~ 0.42 to 0.44 2.8 mins LAR AR ~ 0.36 to 0.38 3.2 mins BLC05 AR = very low 5.2 to 5.6 mins . . It looks like I won't get to find out how the 1060s do under Ryzen, I seem to have bought a dud. ... Stephen :( |
©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.