Message boards :
Number crunching :
2 issues with nvidia gpus
Message board moderation
Author | Message |
---|---|
Bryan Wallace Send message Joined: 28 Jul 00 Posts: 22 Credit: 52,559,173 RAC: 0 |
1. just installed a secondary gpu in ip35 pro rig. seti@home recognizes the gpu as gpu1 (gpu0 is a gtx 260), but in parenthesis says (not used). how do I fix this? 2. on the same rig, my gtx 260 usually takes around 2hrs to finish wu's. I installed the latest lunatics optimized apps and fred's rescheduler, many wu's still take ~ 2hrs but a few only take 20-25 minutes. is this bad luck or is there a problem with the wu's/card? |
Claggy Send message Joined: 5 Jul 99 Posts: 4654 Credit: 47,537,079 RAC: 4 |
1. just installed a secondary gpu in ip35 pro rig. seti@home recognizes the gpu as gpu1 (gpu0 is a gtx 260), but in parenthesis says (not used). how do I fix this? 1. Later versions of Boinc 6.10.x only use the most capable GPU by default, so you'll need a cc_config.xml with the following in it: <cc_config> <options> <use_all_gpus>1</use_all_gpus> </options> </cc_config> You'll need to make this file with notepad, saving it with ANSI encoding, then drop it in your Boinc Data folder (which will be hidden), then restart Boinc. 2. your GTX 260 should take under 20 minutes for most Wu's, with VLAR's being the exception, they'll take up to 2 hours, Claggy |
Bryan Wallace Send message Joined: 28 Jul 00 Posts: 22 Credit: 52,559,173 RAC: 0 |
thank you, will edit the file tonight for dual gpus. looks like my gtx 260 has been stuck with lots of vlars, but it is still getting them even after I ran rescheduler. I'll check that out again tonight. If it's still doing it should I just manually abort all the gpu tasks and allow it to "start over" with new tasks, then run rescheduler right away? |
Claggy Send message Joined: 5 Jul 99 Posts: 4654 Credit: 47,537,079 RAC: 4 |
Something's not quite right with your GTX 260, is there anything else running on your PC with a higher priority than the Cuda app? The tasks i looked at are normal AR tasks, that have taken huge amount of time to complete, all i can think of is that something's restricting the feeder thread from feeding your GPU, i've run into it a few times with my Hybid Nvidia/ATI PC, some ATI apps have caused different Cuda apps to slow to a crawl, Claggy |
Bryan Wallace Send message Joined: 28 Jul 00 Posts: 22 Credit: 52,559,173 RAC: 0 |
there shouldn't be anything at all running on the computer other than a screensaver and background process. the only time it's ever used these days is when my daughter wants to go to dora or diego sites and game, but she can't do that without my help so I would know if she was messing with it. how do I edit priority on the cuda app? |
BilBg Send message Joined: 27 May 07 Posts: 3720 Credit: 9,385,827 RAC: 0 |
there shouldn't be anything at all running on the computer other than a screensaver and background process. the only time it's ever used these days is when my daughter wants to go to dora or diego sites and game, but she can't do that without my help so I would know if she was messing with it. Why do you use a screensaver? CUDA computing + any 3D screensaver (including BOINC screensaver) is not a good idea. Set screensaver to "<None>" and use just option to sleep the Monitor after e.g. 15-20 minutes. If you want from time to time to look at the screensaver make shortcut to it for manual starting: Go to Windows directory Find boinc.scr Make shortcut to it (Send To > Desktop) Set the screensaver to go Blank after 5-10 minutes (just in case). What do you mean by "background process"? Do you mean SETI apps or something else? You do not need to "edit priority on the cuda app" - just make sure that GPU is free - e.g. do not run for a long time 3D screensavers and exit any 3D game or Flash game (browser) when you go to bed. looks like my gtx 260 has been stuck with lots of vlars, but it is still getting them even after I ran rescheduler. I'll check that out again tonight. If it's still doing it should I just manually abort all the gpu tasks and allow it to "start over" with new tasks, then run rescheduler right away? Do you have selected in the rescheduler "Always move VLARs to CPU"? You can also change the "0" (zero) values to something that you want for the particular computer, e.g.: You (and all of CUDA hosts) will continue to receive VLARs in next months so run the rescheduler regularly. To "abort all the gpu tasks and allow it to "start over" with new tasks" is like you want to cut your fingers when you just need to cut your nails. Â - ALF - "Find out what you don't do well ..... then don't do it!" :) Â |
Bryan Wallace Send message Joined: 28 Jul 00 Posts: 22 Credit: 52,559,173 RAC: 0 |
ok, no screensaver is working, 13-17 min for gpu tasks. as mentioned am already running rescheduler every 24 hrs so that should prevent dumping vlars on gpus. now will engage gpu1 and should start to get some good ppd out of this rig! |
Bryan Wallace Send message Joined: 28 Jul 00 Posts: 22 Credit: 52,559,173 RAC: 0 |
thanks for all the help guys! both gpus working flawlessly, taking ~ 15 min for the gtx 260 and 30 min for the 9600gso at present, each has done quite a few wu's since update as well. |
ivan Send message Joined: 5 Mar 01 Posts: 783 Credit: 348,560,338 RAC: 223 |
thanks for all the help guys! both gpus working flawlessly, taking ~ 15 min for the gtx 260 and 30 min for the 9600gso at present, each has done quite a few wu's since update as well. It's worth remembering why screensavers were introduced, and why they may not be necessary these days... Way Back When, displays were CRTs and prolonged electron-beam impact on the phosphors caused a slow reduction in the light- emitting efficiency of the phosphor; this meant that leaving a static display led to a darkening of the permanently-lit parts of the screen -- the image was "burnt in". So, the idea was to detect long periods of inactivity and introduce a dynamic picture to even out the degradation of the phosphors. This wasn't always done well and I've seen cases, ranging from VMS to Linux, where the fancy screensaver stole most if not all of the workstation CPU cycles (we had a Grid admin in our office once whose PC used so much compute power for certain cycled screensavers that the CPU fan would cut in at full speed and drown out conversation...). Modern LCD panels don't have the phosphor burn-in problem of CRTs and don't need a traditional screensaver. A simple blanking command is available which has the benefit of reducing power (to the backlight as well as the LCD) and doesn't run the risk of stealing CPU cycles from any background processes. (Ironically, the above-mentioned admin's panel also blanked, but the screen- saver still cut in even though the display was black; I had to move his mouse to bring up the re-login screen to stop the noise.) |
©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.