Slow gpu comlpetion times

Message boards : Number crunching : Slow gpu comlpetion times

Author	Message
MadMaC Volunteer tester Send message Joined: 4 Apr 01 Posts: 201 Credit: 44,449,547 RAC: 60,180	Message 1069036 - Posted: 21 Jan 2011, 16:52:33 UTC
	Im seeing some really slow completion times on my fermi's, have been since last night. look at the gpu usage - averaging 10% with 3 wu's per card.. It's the same if I drop to one wu per card Have I just got a dodgy batch or is anyone seeing this??? ____________
	ID: 1069036 ·

jravin Send message Joined: 25 Mar 02 Posts: 717 Credit: 61,419,768 RAC: 70,404	Message 1069043 - Posted: 21 Jan 2011, 17:08:30 UTC - in response to Message 1069036.
	Are these WUs VLARs? They run much slower on non-Fermi cards. Perhaps this is also true on Fermi? Anyone out there? Buehller? ____________
	ID: 1069043 ·

Kevin Olley Send message Joined: 3 Aug 99 Posts: 368 Credit: 33,622,151 RAC: 5,933	Message 1069044 - Posted: 21 Jan 2011, 17:08:48 UTC
	Yep, had some slow ones on mine yesterday, looking at the names they are from the same bunch of work units. ____________ Kevin
	ID: 1069044 ·

Kevin Olley Send message Joined: 3 Aug 99 Posts: 368 Credit: 33,622,151 RAC: 5,933	Message 1069047 - Posted: 21 Jan 2011, 17:11:48 UTC - in response to Message 1069043.
	Are these WUs VLARs? They run much slower on non-Fermi cards. Perhaps this is also true on Fermi? Anyone out there? Buehller? Not VLAR's, I am using BoincRescheduler, but just a bunch of slow WU's. ____________ Kevin
	ID: 1069047 ·

[seti.international] Philip J. Fry Volunteer tester Send message Joined: 6 Apr 07 Posts: 6571 Credit: 50,819,577 RAC: 20,436	Message 1069054 - Posted: 21 Jan 2011, 17:22:52 UTC - in response to Message 1069043. Last modified: 21 Jan 2011, 17:24:08 UTC
	Are these WUs VLARs? They run much slower on non-Fermi cards. Perhaps this is also true on Fermi? Anyone out there? Buehller? AFAIK, the x32f CUDA app of Lunatics Installer V0.37 should let run VLAR WUs ltttle bit faster (at least on Fermi) (AFAIK but still long, IIRC GTX480 OC ~ 1 hour). But, usage of BOINC Rescheduler still recommended because of old re-sent VLAR WUs (without .vlar in the name). ____________ >Das Deutsche Cafe. The German Cafe.<
	ID: 1069054 ·

Kevin Olley Send message Joined: 3 Aug 99 Posts: 368 Credit: 33,622,151 RAC: 5,933	Message 1069056 - Posted: 21 Jan 2011, 17:27:06 UTC - in response to Message 1069051.
	I just had to reschedule about 7 VLARs outta the queue on the Frozen 920. Not that it couldn't do them, of course, just wanna avoid slowing the rig down. It's also doing 4 AP WUs right now......amazed it's RAC is holding up so well. Been seeing the odd unmarked VLAR, mostly old ones, but these WU's seemed almost as slow as them but did not identify with BR as such, only noticed them as they were in the waiting upload queue with longer elapsed times than other GPU WU's. Kevin ____________ Kevin
	ID: 1069056 ·

[seti.international] Philip J. Fry Volunteer tester Send message Joined: 6 Apr 07 Posts: 6571 Credit: 50,819,577 RAC: 20,436	Message 1069059 - Posted: 21 Jan 2011, 17:36:23 UTC - in response to Message 1069056. Last modified: 21 Jan 2011, 17:44:42 UTC
	I just had to reschedule about 7 VLARs outta the queue on the Frozen 920. Not that it couldn't do them, of course, just wanna avoid slowing the rig down. It's also doing 4 AP WUs right now......amazed it's RAC is holding up so well. Been seeing the odd unmarked VLAR, mostly old ones, but these WU's seemed almost as slow as them but did not identify with BR as such, only noticed them as they were in the waiting upload queue with longer elapsed times than other GPU WU's. Kevin If the WUs are still on your PC, take (US/UK OS: NotePad, german OS: Editor) and open the WU in question in the setiathome.berkeley.edu project folder, search to angle_range. Maybe the 0.12 GPU VLAR value isn't so well..
	ID: 1069059 ·

[seti.international] Philip J. Fry Volunteer tester Send message Joined: 6 Apr 07 Posts: 6571 Credit: 50,819,577 RAC: 20,436	Message 1069064 - Posted: 21 Jan 2011, 17:44:14 UTC - in response to Message 1069059.
	Is this not longer possible, you could take time and search the results in the tasks overview of your host here online.. ;-) ____________ >Das Deutsche Cafe. The German Cafe.<
	ID: 1069064 ·

Kevin Olley Send message Joined: 3 Aug 99 Posts: 368 Credit: 33,622,151 RAC: 5,933	Message 1069072 - Posted: 21 Jan 2011, 18:12:09 UTC - in response to Message 1069064.
	Is this not longer possible, you could take time and search the results in the tasks overview of your host here online.. ;-) Time is a precious comodity, if resuts are sill available after work I shall have a quick look. Kevin ____________ Kevin
	ID: 1069072 ·

MadMaC Volunteer tester Send message Joined: 4 Apr 01 Posts: 201 Credit: 44,449,547 RAC: 60,180	Message 1069094 - Posted: 21 Jan 2011, 20:01:25 UTC
	Freds rescheduler app isn't seeing them as vlars, maybe it is just a bunch of slow ones.. As Mark said, someone has to crunch them, just wish it wasn't me... ____________
	ID: 1069094 ·

[seti.international] Philip J. Fry Volunteer tester Send message Joined: 6 Apr 07 Posts: 6571 Credit: 50,819,577 RAC: 20,436	Message 1069096 - Posted: 21 Jan 2011, 20:06:49 UTC - in response to Message 1069072. Last modified: 21 Jan 2011, 20:14:06 UTC
	Is this not longer possible, you could take time and search the results in the tasks overview of your host here online.. ;-) Time is a precious comodity, if resuts are sill available after work I shall have a quick look. Kevin I looked TV and had little time.. ;-) I found two very similar ARs: http://setiathome.berkeley.edu/result.php?resultid=1781782752 sent 17 Jan 2011 0:37:58 UTC - received 21 Jan 2011 14:56:06 UTC 0.405797 - 1,642.10 secs Flopcounter: 34298695434430.715000 Spike count: 0 Pulse count: 2 Triplet count: 0 Gaussian count: 0 http://setiathome.berkeley.edu/result.php?resultid=1781782749 sent 17 Jan 2011 0:37:58 UTC - received 21 Jan 2011 14:56:06 UTC 0.406630 - 3,375.75 secs Flopcounter: 34406352504950.312000 Spike count: 10 Pulse count: 2 Triplet count: 1 Gaussian count: 0 This look strange. You let run always same number WUs/GPU? (3?) Maybe the spike, pulse, triplet, gaussian found influence/d the calculation time? Or different ARs simultaneously on GPU? (this could answer maybe only someone of the opt. crew ;-) Maybe unstable OC and the grafic card clocked down, or to idle speed? ____________ >Das Deutsche Cafe. The German Cafe.<
	ID: 1069096 ·

SciManStev Volunteer tester Send message Joined: 20 Jun 99 Posts: 4297 Credit: 63,081,934 RAC: 73,315	Message 1069116 - Posted: 21 Jan 2011, 21:42:20 UTC
	That does seem strange. Even vlar's will show heavy GPU usage, and will actually crunch faster on a Fermi GPU than on a CPU. Your screen is not very responsive though. I would try doing a cold reboot, and see if that fixes it. you might want to see what GPU-Z says about the speed. You may need more voltage, but the very low usage still has me wondering. Steve ____________ Warning, addicted to SETI crunching! Crunching as a member of GPU Users Group. GPUUG Website
	ID: 1069116 ·

SciManStev Volunteer tester Send message Joined: 20 Jun 99 Posts: 4297 Credit: 63,081,934 RAC: 73,315	Message 1069117 - Posted: 21 Jan 2011, 21:43:22 UTC
	You could also run Task Manager, and see if something else is hogging resources. Steve ____________ Warning, addicted to SETI crunching! Crunching as a member of GPU Users Group. GPUUG Website
	ID: 1069117 ·

jravin Send message Joined: 25 Mar 02 Posts: 717 Credit: 61,419,768 RAC: 70,404	Message 1069250 - Posted: 22 Jan 2011, 2:27:45 UTC Last modified: 22 Jan 2011, 2:29:03 UTC
	I just noticed some GPU WUs taking about 1 hour on my GT 240s; normal WUs are anywhere from 20-40 minutes. If it helps, they are all named: 06no10ae.19526.6611.15.10.nnn I get the first part (date and "tape" - ae). But what does the rest signify? (EDIT) and the GPU utilizations seem somewhat lower on these, but not a lot - maybe 20% at the most, (/EDIT) ____________
	ID: 1069250 ·

[seti.international] Philip J. Fry Volunteer tester Send message Joined: 6 Apr 07 Posts: 6571 Credit: 50,819,577 RAC: 20,436	Message 1069251 - Posted: 22 Jan 2011, 2:36:38 UTC - in response to Message 1069250. Last modified: 22 Jan 2011, 2:37:24 UTC
	If the WU is still in BOINC, if you like search the WU in the project folder and look to the AR of the WU (Message 1069059). Use the Windows search function in the project folder. ____________ >Das Deutsche Cafe. The German Cafe.<
	ID: 1069251 ·

Kevin Olley Send message Joined: 3 Aug 99 Posts: 368 Credit: 33,622,151 RAC: 5,933	Message 1069396 - Posted: 22 Jan 2011, 7:28:27 UTC - in response to Message 1069116.
	That does seem strange. Even vlar's will show heavy GPU usage, and will actually crunch faster on a Fermi GPU than on a CPU. Your screen is not very responsive though. I would try doing a cold reboot, and see if that fixes it. you might want to see what GPU-Z says about the speed. You may need more voltage, but the very low usage still has me wondering. Steve Speed is stable and fixed, Asus smartdoctor manual settings. engine 697 MHz, memory 3348 MHz, processor 1394 MHz, voltage 1.025. Same settings for over a week, no unaccounted for errors or invalids. Temps are well within limits 60 - 75 C dependent on card position room temp etc. Running 3 WU's per GPU. CPU is running a mix of Climate Prediction, Einstein and SETI VLAR's, all 4 cores used, was thinking about reducing to 3 cores because of increased processing times on CPU (from when I was using a single 9800GTX+ GPU) but have not tried it yet. Only noticed cause units had not been uploaded, computer had not been used for any other tasks, I had been at work. Workunits with extended times were mainly 28jn10ac and 15ap10aa, Same ranges as OP. I just assumed they were almost but not quite VLAR's. Will give this machine a restart then its bedtime. Kevin ____________ Kevin
	ID: 1069396 ·

Richard Haselgrove Volunteer tester Send message Joined: 4 Jul 99 Posts: 7462 Credit: 39,756,347 RAC: 33,840	Message 1069423 - Posted: 22 Jan 2011, 10:11:59 UTC - in response to Message 1069251.
	If the WU is still in BOINC, if you like search the WU in the project folder and look to the AR of the WU (Message 1069059). Use the Windows search function in the project folder. Or just look on this website. I found 1777833818, 1777833800, 1777833799 on the second page I tried. AR=0.275245 in all cases, so nowhere near VLAR by any definition.
	ID: 1069423 ·

1 · 2 · Next

Message boards : Number crunching : Slow gpu comlpetion times