WARNING: This website is obsolete! Please follow this link to get to the new Albert@Home website!
Binary Radio Pulsar Search (Perseus Arm Survey) v1.33 feedback thread |
Message boards :
Problems and Bug Reports :
Binary Radio Pulsar Search (Perseus Arm Survey) v1.33 feedback thread
Message board moderation
Author | Message |
---|---|
tolafoph Send message Joined: 18 Nov 11 Posts: 3 Credit: 342,780 RAC: 0 |
Hi, I got mir first Binary Radio Pulsar Search (Perseus Arm Survey) tasks and let one run alone on my GTX580. It took 9100 s to run. The old v1.33 tasks take 840 s. I got from GPU-Z: RAM used : min 614 MB , max 718 MB, avg 697 MB GPU loag : min 77 %, max 88 % , avg 86 % This are same values I get for the old BRP4 app. It is the same app version v1.33 , so I should've expected that. I ran two of the new tasks for a few minutes which resulted in the following numbers. RAM used : min 911 MB , max 942 MB, avg 929 MB GPU loag : min 88 %, max 97 % , avg 96 % Will the tasks be this big over at Einstein@home or will you change this? (BRP5 9100 s)/(BRP4 840 s) = 10.8 times longer running times. |
Neil Newell Send message Joined: 9 Jan 13 Posts: 13 Credit: 4,081,564 RAC: 0 |
Seem to have done a few of these too - ~42,000s on 9600GT/8800GT and ~34,000s on HD6850, which is about the same 10x increase in runtime over BRP as you've seen. The tasks that have validated got 2,400 credit each. |
Bikeman (Heinz-Bernd Eggenstein) Volunteer moderator Project administrator Project developer Send message Joined: 28 Aug 06 Posts: 1483 Credit: 1,864,017 RAC: 0 |
Yes, the plan is to do it like that on E@H as well. The rather short run times (for E@H standards) of the BRP4 GPU tasks were the result of BRP (formerly known as ABP) being a CPU app first which then gradually was optimized for GPUs. At first the speedups were moderate (a factor of 2...3) but with more and more parts of the code being ported to the GPU, the performance gap between CPU and GPU versions got wider and wider...to a point where it finally became difficult to have the same tasks for CPUs and GPUs. So for the new search, we hope to be able to find more pulsars by investing more time into each individual WU, and not limit ourselves to run times that would be tailored to be acceptable for CPU hosts. Cheers HB |
zombie67 [MM] Send message Joined: 10 Oct 06 Posts: 130 Credit: 30,924,459 RAC: 0 |
BOINCtasks is reporting that these do not checkpoint. Is this correct? Edit: Actually, BOINCtasks is showing something very weird. The run time is 8h35m, and BT is showing the last checkpoint was 15h00m ago. Is there something funky about this app that might cause that? Dublin, California Team: SETI.USA |
zombie67 [MM] Send message Joined: 10 Oct 06 Posts: 130 Credit: 30,924,459 RAC: 0 |
FWIW, this is the task: http://albert.phys.uwm.edu/result.php?resultid=831638 Dublin, California Team: SETI.USA |
skgiven Send message Joined: 14 Oct 12 Posts: 9 Credit: 4,734,887 RAC: 0 |
Just an oddity I spotted in a tasks properties: CPU time at last checkpoint 02:51:58 CPU time 01:44:51 Elapsed time 03:50:40 With most projects it's usually the case that the last checkpoint time is derived from CPU time, but perhaps it's being taken from Elapsed time here? https://albert.phys.uwm.edu/result.php?resultid=1019875 |
skgiven Send message Joined: 14 Oct 12 Posts: 9 Credit: 4,734,887 RAC: 0 |
Was running a Binary Radio Pulsar Search (Perseus Arm Survey) v1.39 (BRP5-opencl-ati) on an HD5850 (just one task). It's progress was only 2% after 1.5h. I checked GPUZ and found that the GPU wasn't being used, just the occasional blip. The GDDR was jumping back and forth from 300MHz to 1000MHz. I tried to suspend and resume but it's behavior was the same. I aborted the WU: https://albert.phys.uwm.edu/result.php?resultid=1029977 |