Message boards : News : New systems in long queue
| Author | Message |
|---|---|
|
Hi guys, | |
ID: 27062 | Rating: 0 | rate:
 /   Reply
Quote
| |
 skgivenVolunteer moderator Volunteer tester  Send message Joined: 23 Apr 09 Posts: 3968 Credit: 1,995,359,260 RAC: 0 Level  Scientific publications       | |
|
New research? | |
| ID: 27063 | Rating: 0 | rate:
/ Reply
Quote
| |
|
It is actually the continuation of a system I sent last month (domain SH2), and worked quite well, but I need more data :) | |
| ID: 27064 | Rating: 0 | rate:
/ Reply
Quote
| |
|
and we'll be happy to oblige :) | |
| ID: 27065 | Rating: 0 | rate:
/ Reply
Quote
| |
Just sent a good amount of WUs to the long queue... I've been fishing for these for nearly 3 days and have come up empty...<sniff, sniff> :'( ____________ | |
| ID: 27066 | Rating: 0 | rate:
/ Reply
Quote
| |
|
TheFiend Send message Joined: 26 Aug 11 Posts: 99 Credit: 2,500,112,138 RAC: 0 Level  Scientific publications  | |
Just sent a good amount of WUs to the long queue... Just caught my first one...... looks to be around 8 hours on my GTX670 | |
| ID: 27080 | Rating: 0 | rate:
/ Reply
Quote
| |
Just caught my first one...... looks to be around 8 hours on my GTX670 I finally caught one for my GTX 570. It's in line behind a NATHAN that won't finish for about 8 hours, though. ____________ | |
| ID: 27082 | Rating: 0 | rate:
/ Reply
Quote
| |
|
Caught one too, (ala_structure_191-NOELIA_1CPU_sh2fragment-0-3-RND2901) | |
| ID: 27083 | Rating: 0 | rate:
/ Reply
Quote
| |
|
TheFiend Send message Joined: 26 Aug 11 Posts: 99 Credit: 2,500,112,138 RAC: 0 Level Scientific publications | |
Just sent a good amount of WUs to the long queue... Finished in 8.6 hours on my GTX670 and scored 114300, also used less CPU time!!! http://www.gpugrid.net/result.php?resultid=5923417 | |
| ID: 27084 | Rating: 0 | rate:
/ Reply
Quote
| |
Finished in 8.6 hours on my GTX670 and scored 114300... About 11.6 hours for this one, 3761726, on my stock GTX 570. Good for 95,250. Crunching my second one now. :-) ____________ | |
| ID: 27088 | Rating: 0 | rate:
/ Reply
Quote
| |
|
Hi there! | |
| ID: 27151 | Rating: 0 | rate:
/ Reply
Quote
| |
Carlesa25 Send message Joined: 13 Nov 10 Posts: 328 Credit: 72,619,453 RAC: 0 Level  Scientific publications | |
|
Hello Zac: I think it best to cancel the task and downloads only short drives - ACEMD standard-which are also very valuable work. | |
| ID: 27154 | Rating: 0 | rate:
/ Reply
Quote
| |
|
Granite T. Rock Send message Joined: 7 Sep 09 Posts: 3 Credit: 7,778,853 RAC: 0 Level  Scientific publications | |
|
Generally I thought BOINC downloaded work units based upon if it thinks you can finish it on time? Or is that something implemented server side? I only take a general interest in my work units once every couple of months. Hate to think people's cycles are being wasted because the clients are downloading more than they can chew. | |
| ID: 27155 | Rating: 0 | rate:
/ Reply
Quote
| |
|
skgiven Volunteer moderator Volunteer tester Send message Joined: 23 Apr 09 Posts: 3968 Credit: 1,995,359,260 RAC: 0 Level Scientific publications | |
|
The Quadro NVS 4200M (GF119) only has 48 cuda cores (shaders), of which only 32 would be used (CC2.1). It's also GDDR3 and only 155GFlops peak, so I doubt that it would finish any task in time. I would not recommend using that GPU here. If you still want to use it I suggest you try another project (it might work on POEM). | |
| ID: 27165 | Rating: 0 | rate:
/ Reply
Quote
| |
|
[AF>Belgique] bill1170 Send message Joined: 4 Jan 09 Posts: 13 Credit: 1,292,573,895 RAC: 4,154,165 Level  Scientific publications | |
|
Dear Noelia, | |
| ID: 27670 | Rating: 0 | rate:
/ Reply
Quote
| |
|
Dear volunteers, | |
| ID: 27672 | Rating: 0 | rate:
/ Reply
Quote
| |
Dear volunteers, Thanks a lot ! ____________ Lubuntu 16.04.1 LTS x64 | |
| ID: 27674 | Rating: 0 | rate:
/ Reply
Quote
| |
|
skgiven Volunteer moderator Volunteer tester Send message Joined: 23 Apr 09 Posts: 3968 Credit: 1,995,359,260 RAC: 0 Level Scientific publications | |
Dear volunteers, Does this mean we should abort these tasks? I was sent two sh2forces tasks yesterday evening/overnight. They are now at >65% and 92%. Just under 2h left on one! Anyway, I did some testing on them. To cut a long story short, the performance is directly limited by the tasks CPU requirement. A solution could be going multi-threaded (mt), running some of those CPU cycles on the GPU (if possible), or running multiple WU's on the same GPU. Beta testing would be a good idea too! The slowness isn't due to PCIE bandwidth, I tested at PCIE3.0 x16 and PCIE2.0 x8; no difference in performance. It's also not a controller bottleneck, or too much memory being used. It's just down to the CPU usage and CPU to GPU performance ratio. A single thread from a fast CPU isn't capable of supporting the requirements of a low to mid range GPU. Even a GTX460 supported by a highly overclocked i7 would not yield good GPU utilization: By crunching fewer CPU tasks I was able to improve performance, somewhat. (i7-3770K@4.2GHz HT on, 8GB DDR3-2133, SATA6 SSD, W7x64) When CPU usage was at 98% (crunching CPU tasks on 6 threads + 2 GPU tasks) GPU Load - GTX660Ti 7% to 10% - GTX470 10% Mem Ctrl - GTX660Ti 3% - GTX470 2% When CPU usage was at 46% (crunching CPU tasks on 2 threads + 2 GPU tasks) GPU Load - GTX660Ti 9% to 12% - GTX470 11% to 15% Mem Ctrl - GTX660Ti 4% - GTX470 3% When CPU usage was at 21% (only crunching 2 GPU tasks) GPU Load - GTX660Ti 12% to 14% - GTX470 15% to 17% Mem Ctrl - GTX660Ti 5% - GTX470 3% With HT off and CPU usage was at 43% (only crunching 2 GPU tasks) GPU Load - GTX660Ti 13% - GTX470 17% Mem Ctrl - GTX660Ti 5% - GTX470 4% The GPU temperatures and fan speeds rose after freeing CPU cores and turning HT off (not that I would normally recommend turning it off). Even with such optimizations however the GTX660Ti was only using 35% of it's power consumption capabilities. ____________ FAQ's HOW TO: - Opt out of Beta Tests - Ask for Help | |
| ID: 27682 | Rating: 0 | rate:
/ Reply
Quote
| |
Retvari Zoltan Send message Joined: 20 Jan 09 Posts: 2343 Credit: 16,206,655,749 RAC: 261,147 Level  Scientific publications | |
Batch group "sh2forces" requires extensive CPU usage and that's why some of you have noticed a big slow down. Same problem as we were having with previous systems: Strange. Very strange. You didn't said who aborted these workunits... It seems that you didn't successfully aborted this batch, so the crunchers have do it one by one. | |
| ID: 27683 | Rating: 0 | rate:
/ Reply
Quote
| |
|
klepel Send message Joined: 23 Dec 09 Posts: 189 Credit: 4,718,545,137 RAC: 2,198,542 Level  Scientific publications | |
|
Noelia, If these work unites are important for your work I will crunch all 100 WUs "sh2forces" happily. Isn’t it about science? As I have written before, yes credits are important for me as well, but if everybody gets one or two of these slow not optimal WUs to crunch, it affects all and nobody gets really hurt credit wise. | |
| ID: 27694 | Rating: 0 | rate:
/ Reply
Quote
| |
|
HA-SOFT, s.r.o. Send message Joined: 3 Oct 11 Posts: 100 Credit: 5,879,292,399 RAC: 0 Level  Scientific publications | |
|
No problem for me also. | |
| ID: 27696 | Rating: 0 | rate:
/ Reply
Quote
| |
|
Retvari Zoltan Send message Joined: 20 Jan 09 Posts: 2343 Credit: 16,206,655,749 RAC: 261,147 Level Scientific publications | |
No problem for me also. The performance of these workunits on Linux is ~3 times better than on Windows. There is no much difference between XP or 7 regarding these workunits. I've aborted another one, it was 12% complete after 5 hours. It's not so popular. I've suspended every CPU task to see if there's any improvement in GPU usage, and my GPU usage's gone up by 3% to 11%. I still consider these workunits as a waste of resources (on Windows). | |
| ID: 27697 | Rating: 0 | rate:
/ Reply
Quote
| |
|
skgiven Volunteer moderator Volunteer tester Send message Joined: 23 Apr 09 Posts: 3968 Credit: 1,995,359,260 RAC: 0 Level Scientific publications | |
No problem for me also. Linux GTX 680: Run time 44,471.89 CPU time 38,477.50 W7 (GTX660Ti, GTX470): ala_structure_881-NOELIA_sh2forces-3-5-RND6660_1 3934692 139265 12 Dec 2012 | 0:34:09 UTC 13 Dec 2012 | 7:08:28 UTC Completed and validated 86,697.19 78,124.14 79,375.00 Long runs (8-12 hours on fastest card) v6.16 (cuda42) ala_structure_241-NOELIA_sh2forces-1-5-RND3374_3 3931355 139265 11 Dec 2012 | 19:34:19 UTC 13 Dec 2012 | 1:47:29 UTC Completed and validated 88,036.52 79,443.15 79,375.00 Long runs (8-12 hours on fastest card) v6.16 (cuda42) I don't know what else (HA-SOFT, s.r.o.)'s Linux system was doing, if anything, (possibly some WCG or Rosetta CPU tasks) but if it wasn't I would say it's roughly twice as fast as W7. If it was crunching CPU tasks then I would agree it could be close to 3 times as fast. My tasks were initially run while crunching CPU tasks. Around half way through I optimized for the GPUGrid tasks. The limiting factor on Windows appears to be the CPU requirement, hence the GTX470 was almost as fast as the GTX660Ti. Even on Linux, and assuming it's 3times as fast, GPU utilization would only be ~39%. If the research is so important, make it Linux only, or better still, make an mt app and feed the GPU properly. ____________ FAQ's HOW TO: - Opt out of Beta Tests - Ask for Help | |
| ID: 27698 | Rating: 0 | rate:
/ Reply
Quote
| |
|
HA-SOFT, s.r.o. Send message Joined: 3 Oct 11 Posts: 100 Credit: 5,879,292,399 RAC: 0 Level Scientific publications | |
6 HT Cores WCG 2 HT Cores left for GPUGRID | |
| ID: 27699 | Rating: 0 | rate:
/ Reply
Quote
| |
|
skgiven Volunteer moderator Volunteer tester Send message Joined: 23 Apr 09 Posts: 3968 Credit: 1,995,359,260 RAC: 0 Level Scientific publications | |
|
Thanks, ~3times as fast it is then ;) | |
| ID: 27700 | Rating: 0 | rate:
/ Reply
Quote
| |
Message boards : News : New systems in long queue
