Questions and Answers :
GPU applications :
2 different cuda cards
Message board moderation
Author | Message |
---|---|
bloodrain Send message Joined: 8 Dec 08 Posts: 231 Credit: 28,112,547 RAC: 1 |
what my question is . lets say i got a 9800 or 9600gt and a 9800gtx in none sli mode. could it run seti cuda? |
OzzFan Send message Joined: 9 Apr 02 Posts: 15691 Credit: 84,761,841 RAC: 28 |
Yes. |
bloodrain Send message Joined: 8 Dec 08 Posts: 231 Credit: 28,112,547 RAC: 1 |
ok thank you. might want to put that in faqs . very good to know |
Hammeh Send message Joined: 21 May 01 Posts: 135 Credit: 1,143,316 RAC: 0 |
Yes. BOINC Manager should list them all in the messages tab on client startup. |
Questor Send message Joined: 3 Sep 04 Posts: 471 Credit: 230,506,401 RAC: 157 |
Yes. BOINC Manager should list them all in the messages tab on client startup. I had a spare CUDA card sitting idle so decided to put it one of my PCs - now has 2 cards - GE8600GTS and GE9600GT. Windows shows both cards are working OK and BOINC shows two GPUS after I sorted out my cc_config file. Only thing is, BOINC messages shows two separate lines both showing the GE8600GTS card - doesn't mention the 9600. When I look at the machine on the web site it shows 2 times the 9600 card [2] NVIDIA GeForce 9600 GT (511MB) driver: 18585 Is this normal behaviour? I am running 6.6.36 - previously 6.6.33 when the card was installed a couple of days ago. Have never run two cards together before so don't know what to expect. Both seem to be running OK but I have seen other posts in the forums (but for different BOINC versions) where the message tab shows :- CUDA card x, CUDA card y whereas mine shows :- CUDA card x CUDA card x. GPU Users Group |
Ben Panske Send message Joined: 30 May 08 Posts: 3 Credit: 226,602 RAC: 0 |
What changes need to be made to this. I am currently at work and am not infront of my client to see if it is registering both cards correctly. Should it automatically detect both cards and run good off the start or do I need to modify my cc_config file to make it function correctly |
Questor Send message Joined: 3 Sep 04 Posts: 471 Credit: 230,506,401 RAC: 157 |
What changes need to be made to this. I am currently at work and am not infront of my client to see if it is registering both cards correctly. Should it automatically detect both cards and run good off the start or do I need to modify my cc_config file to make it function correctly I now have this :- <cc_config> <options> <use_all_gpus>1</use_all_gpus> </options> </cc_config> Ref use_all_gpus (New in 6.6.25) Without this setting I was only seeing the one GPU. May be because the cards have quite a different spec. At startup the messages tab shows :- 11/06/2009 21:55:10 CUDA device: GeForce 8600 GTS (driver version 18585, compute capability 1.1, 256MB, est. 18GFLOPS) 11/06/2009 21:55:10 CUDA device: GeForce 8600 GTS (driver version 18585, compute capability 1.1, 256MB, est. 18GFLOPS) The new setting may account for it showing 2 * the lower spec card - although odd the web site shows 2 * the higher spec card. I see now from your other post (906265) you are having the same issue as me. GPU Users Group |
Byron S Goodgame Send message Joined: 16 Jan 06 Posts: 1145 Credit: 3,936,993 RAC: 0 |
With 6.6.31 there's info in this message that also talks about an edit to the cc_config.xml file that needs the extra setup switch, but also extending the desktop to the second monitor/dummy VGA plug first, if you haven't already tried this. |
Questor Send message Joined: 3 Sep 04 Posts: 471 Credit: 230,506,401 RAC: 157 |
With 6.6.31 there's info in this message that also talks about an edit to the cc_config.xml file that needs the extra setup switch, but also extending the desktop to the second monitor/dummy VGA plug first, if you haven't already tried this. The dummy plug seems to be to overcome the second card being disabled by Windows? In my case both cards are being used it is just that BOINC reports the cards incorrectly i.e. both 8600 rather than 8600/9600 (although web page shows 2 * 9600). This may be deliberate behaviour of the app as there is only one DCF and if it worked on the basis of the faster card it would download too many GPU tasks which could not complete because one card is too slow but I'm just guessing - there's quite a difference in performance between the two cards (18/37 Gflops). GPU Users Group |
Byron S Goodgame Send message Joined: 16 Jan 06 Posts: 1145 Credit: 3,936,993 RAC: 0 |
It's my understanding the dummy plug/second monitor, is so that the 2nd card can be used. In my case both cards are being used it is just that BOINC reports the cards incorrectly i.e. both 8600 rather than 8600/9600 (although web page shows 2 * 9600). In a few of your tasks I looked into, I notice when the 8600 is used you are getting the message "setiathome_CUDA: CUDA runtime ERROR in plan FFT. Falling back to HOST CPU processing..." and with the other behaviour you mentioned is what lead me to offer what I'd seen in the post I mentioned. |
Questor Send message Joined: 3 Sep 04 Posts: 471 Credit: 230,506,401 RAC: 157 |
I'll check it out. I hadn't spotted the CUDA errors you mentioned. Thanks. GPU Users Group |
Byron S Goodgame Send message Joined: 16 Jan 06 Posts: 1145 Credit: 3,936,993 RAC: 0 |
I guess most of the ones I saw have already been deleted, though resultid=1244080225 still shows the error and was marked invalid in the end. I also notice several task marked "Completed, validation inconclusive" such as resultid=1244074646 and resultid=1242858539 that were done by the 8600, so it could be a sign of a problem, but until more of them are finished by the wingman it's hard to say with any certainty, though it does seem from the few other tasks I've checked, a majority of the tasks marked "Completed, validation inconclusive" are from tasks where a CUDA card aws used. Edit: there are also several tasks marked as valid from the 8600 as well as many from the other cards used. |
©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.