SETI@home v7 6.98 for NVIDIA CUDA 2.3, 3.2, and 4.2 released.
Message boards :
News :
SETI@home v7 6.98 for NVIDIA CUDA 2.3, 3.2, and 4.2 released.
Message board moderation
Previous · 1 . . . 6 · 7 · 8 · 9
| Author | Message |
|---|---|
 Send message Joined: 10 Mar 12 Posts: 1700 Credit: 13,216,373 RAC: 0 
|
It's quite clear that the idea that the system would find out which app (in my case 2.2, 2.3, or 3.2) is the fastest, and in time only send the fastest is not working. I still get the slowest app also, and by looking at the "Application details for host" pages for those computers, it's painfully clear that it sends most WU's for the app with the shortest "Average turnaround time", and not the fastests "Average processing rate". However, it never stops sending apps for any version, not even the app for the longest "Average turnaround time" That's not the way to determine which app is the fastests. There's many reasons why the slowest app, may have shorter "Average turnaround time" than the fastest app. The absolute slowest app for me, 2.2 should by looking at the APR have stopped being sent by now, but it still keeps coming. Also, when looking at my host 57179, its seems as if the APR for 2.2 is faster than 3.2, but that simply is not right. Even though the APR says otherwise, on that host 3.2 is very very much faster than 2.2, and 2.3 is much faster than 2.2 and 3.2. There's something fishy when it comes to the algorithms determining which app is the fastest. Relevant "Application details for host" pages: hostid=57176 hostid=57179 And with that, ends the voting from the Swedish jury :-) Added: Also, come to think of it, another strange thing. When my host 57176 asked for new GPU tasks, it got 8 Cuda23. But due to the congestion of the servers, those never arrived, but became ghosts. At the next request when they finally arrived at my little home as resends, they had been renamed as Cuda22 tasks, the slowest of them all to crunch. Now, the Swedish jury rest its case. ID: 44299 ·  |
|
Send message Joined: 29 May 06 Posts: 1037 Credit: 8,440,339 RAC: 0 
|
My GTX460 now has ~400 completed Cuda32 and Cuda42 results, the server still hasn't decided if one is faster than the other: http://setiweb.ssl.berkeley.edu/beta/host_app_versions.php?hostid=45274 the 9800GTX+ is also still getting a mix of Wu for Cuda22, Cuda23, Cuda32 and Cuda42, according to the APR page, it is preferring the Cuda32 app inspite it not being quite the fastest: http://setiweb.ssl.berkeley.edu/beta/host_app_versions.php?hostid=15616 Claggy ID: 44301 · |
|
Send message Joined: 3 Jan 07 Posts: 1451 Credit: 3,272,268 RAC: 0
|
For a GTX 670 'Kepler', it's quite clear that cuda42 is faster - just watching the tasks go through, and from the application details page: http://setiweb.ssl.berkeley.edu/beta/host_app_versions.php?hostid=59866 From the figures - 433 tasks at APR 180 for cuda32, and 587 tasks at APR 271 for cuda42 - the machine has probably spent about the same amount of time on each application: and look, 24 new tasks today for each of them. But an exclusive diet of cuda42 would be more efficient. ID: 44302 · |
 Eric J Korpela Send message Joined: 15 Mar 05 Posts: 1547 Credit: 27,183,456 RAC: 0 
|
There's something deeply wrong with the version choosing logic in the scheduler. Unfortunately I don't have time to worry about it right now. 
ID: 44303 · |
|
Send message Joined: 29 May 06 Posts: 1037 Credit: 8,440,339 RAC: 0
|
Do you have a forecast for the 6.99 CPU apps, and the 6.99 x41zb Cuda apps? Claggy ID: 44317 · |
|
Eric J Korpela Send message Joined: 15 Mar 05 Posts: 1547 Credit: 27,183,456 RAC: 0
|
As soon as I dig myself out of this avalanche. Hopefully I'll get them released next week.
ID: 44322 · |
|
Send message Joined: 29 May 06 Posts: 1037 Credit: 8,440,339 RAC: 0
|
O.K Thanks Claggy ID: 44324 · |
|
Eric J Korpela Send message Joined: 15 Mar 05 Posts: 1547 Credit: 27,183,456 RAC: 0
|
New scheduler is on. Let me know if you have problems with anything.
ID: 44331 · |
|
Send message Joined: 24 Aug 09 Posts: 79 Credit: 26,117 RAC: 0
|
Any ETA on GPU apps for OSX? I have one standing by when ever it is released. ID: 44340 · |
Raistmer Send message Joined: 18 Aug 05 Posts: 2423 Credit: 15,878,738 RAC: 0 
|
http://setiweb.ssl.berkeley.edu/beta/result.php?resultid=11378318 stderr looks completely OK, but task result "-1" unknown error.... And I have few such results on different hosts. What could be wrong ? Detected setiathome_enhanced_v7 task. Autocorrelations enabled, size 128k elements. ID: 44419 · |
|
Raistmer Send message Joined: 18 Aug 05 Posts: 2423 Credit: 15,878,738 RAC: 0
|
We should admit now that BOINC can't determine faster app. At all. SETI@home v7 6.98 windows_intelx86 (cuda23) Number of tasks completed 147 Max tasks per day 188 Number of tasks today 0 Consecutive valid tasks 155 Average processing rate 128.85971698782 Average turnaround time 5.40 days SETI@home v7 6.98 windows_intelx86 (cuda32) Number of tasks completed 142 Max tasks per day 182 Number of tasks today 0 Consecutive valid tasks 149 Average processing rate 84.86276867162 Average turnaround time 4.53 days SETI@home v7 6.98 windows_intelx86 (cuda22) Number of tasks completed 134 Max tasks per day 55 Number of tasks today 0 Consecutive valid tasks 23 Average processing rate 55.935884802959 Average turnaround time 4.48 days Looks for number of completed results. CUDA22 slowest, but it completed only fractionally less tasks though number of all completed tasks is quite high now, much more than required for this BOINC ability to start working. Hence: it doesn't work. Will try to ping David for this issue too... ID: 44420 · |
|
Eric J Korpela Send message Joined: 15 Mar 05 Posts: 1547 Credit: 27,183,456 RAC: 0
|
Yeah. I agree. I'll put time into finding out why next week.
ID: 44425 · |
|
Send message Joined: 29 May 06 Posts: 1037 Credit: 8,440,339 RAC: 0
|
http://setiweb.ssl.berkeley.edu/beta/result.php?resultid=11378318 That is a bug with the 6.98 Cuda22 app/api, when suspended and resumed, the app/api throws an error, it has been fixed in the 6.99 Cuda22 app. (Initial testing of the Cuda22 apps had this error on every completion, later variants fixed this problem, but the problem with suspend & resuming wasn't spotted until the app went live) Claggy ID: 44432 · |
|
Send message Joined: 29 May 06 Posts: 1037 Credit: 8,440,339 RAC: 0
|
Yeah. I agree. I'll put time into finding out why next week. Another strange thing i noted when the Cuda 6.99 apps were released, on the app version page the entries changed from being for 6.98 to 6.99, then later at least one of them switched back to 6.98 (before the outage) At the moment i have the following, i expect once the 6.99 Cuda32 app version has some validated again, the 6.98 Cuda32 entry will disappear again: Application details for host 45274 SETI@home v7 6.98 windows_intelx86 Number of tasks completed 1002 Max tasks per day 28 Number of tasks today 0 Consecutive valid tasks 0 Average processing rate 18.495731245121 Average turnaround time 7.70 days SETI@home v7 6.98 windows_intelx86 (cuda32) Number of tasks completed 854 Max tasks per day 36 Number of tasks today 0 Consecutive valid tasks 11 Average processing rate 211.1938226449 Average turnaround time 4.58 days SETI@home v7 6.99 windows_intelx86 (ati_opencl_sah) Number of tasks completed 1442 Max tasks per day 79 Number of tasks today 0 Consecutive valid tasks 46 Average processing rate 109.17344183925 Average turnaround time 4.74 days SETI@home v7 6.99 windows_intelx86 (cuda42) Number of tasks completed 937 Max tasks per day 27 Number of tasks today 0 Consecutive valid tasks 3 Average processing rate 232.33883874667 Average turnaround time 6.81 days SETI@home v7 6.99 windows_intelx86 (cuda32) Number of tasks completed 0 Max tasks per day 30 Number of tasks today 0 Consecutive valid tasks 0 Average turnaround time 25.00 days Claggy ID: 44435 · |
|
Send message Joined: 29 May 06 Posts: 1037 Credit: 8,440,339 RAC: 0
|
And today i'm back to having both v7 6.98 (cuda32) and (cuda42) entries, and only the v7 6.99 (cuda32) entry, the v7 6.99 Cuda42 entry having disappeared, Claggy ID: 44476 · |
|
Send message Joined: 3 Jan 07 Posts: 1451 Credit: 3,272,268 RAC: 0
|
I moved a host between accounts three days ago, so 59866 became 61440 (no hardware or software change). This is the GTX 670 'Kepler' which is some 50% more efficient with cuda42 than with cuda32. So far, the new host 61440 has recorded in application details: SETI@home v7 6.99 windows_intelx86 (cuda42) Number of tasks completed 39 Max tasks per day 71 Number of tasks today 2 Consecutive valid tasks 39 Average processing rate 306.62974428217 Average turnaround time 36.19 days SETI@home v7 6.99 windows_intelx86 (cuda32) Number of tasks completed 0 Max tasks per day 138 Number of tasks today 0 Consecutive valid tasks 105 Average turnaround time 0.00 days The '(consecutive) valid tasks' are right - 39 for cuda42, and 105 for cuda32, from the 144 validated - but I can't explain the turnround time of 36.19 days for a three day old host, nor the lack of any completions/APR for cuda32. ID: 44478 · |
|
Send message Joined: 29 May 06 Posts: 1037 Credit: 8,440,339 RAC: 0
|
On my 9800GTX+ host, the applications details page is not showing any completed tasks or APR values for the 6.99 Cuda23 or Cuda32 planclasses (they are the fastest two and they have completed full tasks), but is for the Cuda22 and the Cuda42 planclasses: Application details for host 15616 SETI@home v7 6.99 windows_intelx86 (cuda42) Number of tasks completed 19 Max tasks per day 53 Number of tasks today 0 Consecutive valid tasks 20 Average processing rate 92.05891786671 Average turnaround time 0.86 days SETI@home v7 6.99 windows_intelx86 (cuda32) Number of tasks completed 0 Max tasks per day 49 Number of tasks today 0 Consecutive valid tasks 16 Average turnaround time 0.00 days SETI@home v7 6.99 windows_intelx86 (cuda23) Number of tasks completed 0 Max tasks per day 41 Number of tasks today 0 Consecutive valid tasks 8 Average turnaround time 0.00 days SETI@home v7 6.99 windows_intelx86 (cuda22) Number of tasks completed 11 Max tasks per day 45 Number of tasks today 0 Consecutive valid tasks 12 Average processing rate 58.137014408776 Average turnaround time 1.19 days Claggy ID: 44496 · |
|
Send message Joined: 10 Mar 12 Posts: 1700 Credit: 13,216,373 RAC: 0
|
The '(consecutive) valid tasks' are right - 39 for cuda42, and 105 for cuda32, from the 144 validated - but I can't explain the turnround time of 36.19 days for a three day old host, nor the lack of any completions/APR for cuda32. Same tendency for me, on both my Cuda computers. No completions/APR for any other app than cuda22. I have far fewer numbers completed, but I can not see any outliers amongs those finished WU's from cuda23 or 32 (I don't do Cuda42, since as you all know, I do not upgrade to fix something that isn't broken), still those apps never get any increase in completions/APR. Hmm, this belongs in the 6.99 thread though, both of our posts.... ID: 44497 · |
©2025 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.