Dear Amber users
My name is jong hui hong in Korea
Hi recently We have installed Amber12
and It seems to works quiet well
But in case of multiple GPU card system, our ssystem has a problem in
selecting
a specific GPU device
suppose We have 3 GPU card
[user1.cs-g ~]$ deviceQueryDrv -noprompt | egrep "^Device"
Device 0: "Tesla C2075"
Device 1: "Tesla C2075"
Device 2: "Tesla C2075"
If I wanto select device 0
[user1.cs-g ~]$ export CUDA_VISIBLE_DEVICES="0"
[user1.cs-g ~]$ deviceQueryDrv -noprompt | egrep "^Device"
Device 0: "Tesla C2075"
It;s OK
But If I want select device 1
[user1.cs-g ~]$ export CUDA_VISIBLE_DEVICES="1"
[user1.cs-g ~]$ deviceQueryDrv -noprompt | egrep "^Device"
Device 0: "Tesla C2075"
even I set the CUDA_VISIBLE_DEVICES to 1
It only shows Deice 0
If I want to select device 0,1
[user1.cs-g ~]$ deviceQueryDrv -noprompt | egrep "^Device"
Device 0: "Tesla C2075"
Device 1: "Tesla C2075"
It show both of them
But In case of I want to select device 0,2
[user1.cs-g ~]$ export CUDA_VISIBLE_DEVICES="0,2"
[user1.cs-g ~]$ deviceQueryDrv -noprompt | egrep "^Device"
Device 0: "Tesla C2075"
Device 1: "Tesla C2075"
even I set CUDA_VISIBLE_DEVICES to 0,2
It still shows device 0,1
I think there's somethin's wrong with cuda setup in our system, but I can
not manage at all..
Any kind of suggestion would be reall appreciated.
Best Regards
Jong hui
[?usn=13646389&email=croshong.nate.com&key=dc8a17ea8fa5db6b89fb11119905a9a4$
2d84fe89.mail3.nate.com]
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Wed Nov 21 2012 - 03:00:02 PST
