Advanced search

Message boards : Graphics cards (GPUs) : RDP and CUDA

Author Message
Profile paul
Avatar
Send message
Joined: 16 Oct 08
Posts: 7
Credit: 3,999,754
RAC: 0
Level
Ala
Scientific publications
watwatwatwat
Message 5064 - Posted: 30 Dec 2008 | 2:54:03 UTC
Last modified: 30 Dec 2008 | 3:07:30 UTC

said by Neil A in another thread-

I access one of my home crunching computers a lot by MS Terminal Services. I have a feeling that if I access the computer remotely from my work laptop, that when it finishes completing the current work unit on my CPU (8800GT) that the next one doesn't start correctly and errors out. IF it connect to the computer directly at its console again before the work unit completes, then the next work unit starts. This is a theory at this point only.

I can only guess that since I have connected from a laptop that doesn't have a valid GPU to process this type of work, and if I remain connected to it, that when the current WU completes and the next WU starts, it sees the GPU card on my laptop instead and errs out?

This is only a guess on my part, and I'm still trying to prove this theory. Does anybody else access their crunching home computer a lot via terminal services and see this behavior?

Does anybody know what should happen to a crunching computer if you access it remotely from a non-compatible laptop via terminal services?


I'm seeing this as well, check the rig in the office with RDP, and it starts dumping GPUGrid work. Frustrating, as it's hard enough keeping the 260 with enough work. Running 180.60 CUDA, NVidia 180.84, BOINC 6.5.0 and XP64.

Xaak explained it to me yesterday in our IRC channel what was going on, suggested Dameware as an alternative. Thought this should have it's own thread for specific answers or explanations.
____________
Team Starfire World BOINC
IRC- irc//irc.teamstarfire.net:6667/team_starfire

ExtraTerrestrial Apes
Volunteer moderator
Volunteer tester
Avatar
Send message
Joined: 17 Aug 08
Posts: 2705
Credit: 1,311,122,549
RAC: 0
Level
Met
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 5152 - Posted: 1 Jan 2009 | 17:23:29 UTC

Ah, I think I was seeing the same behaviour with S@H! The 1st completed and all other ones just crashed. I thought it probably was their app (and went into holiday) .. but this explanation seems much more likely.

MrS
____________
Scanning for our furry friends since Jan 2002

Neil A
Send message
Joined: 9 Oct 08
Posts: 50
Credit: 12,676,739
RAC: 0
Level
Pro
Scientific publications
watwatwatwatwatwatwatwat
Message 5179 - Posted: 3 Jan 2009 | 1:30:56 UTC
Last modified: 3 Jan 2009 | 1:39:37 UTC

I looked in the NVIDIA CUDA forum regarding various posts and found one that explained that when you terminal service from one non-cuda PC to a CUDA pc running GPU work and remain connected when the current task ends and tries to start a new one, it will fail. This is a result of the terminal services session setting itself up with a non cuda driver/configuration so GPU work fails. Seems silly to me when GPU capable graphics cards on the CUDA capable PC, but this is the way the application works.

A suggestion was to use VNC or similar alternate product to perform remote connections to CUDA pc's. I'll look for the post again on the NVIDIA CUDA forum and post it here for those who may be interested.

Here it is: link

Neil
____________
Crunching for the benefit of humanity and in memory of my dad and other family members.

Profile paul
Avatar
Send message
Joined: 16 Oct 08
Posts: 7
Credit: 3,999,754
RAC: 0
Level
Ala
Scientific publications
watwatwatwat
Message 5182 - Posted: 3 Jan 2009 | 5:03:05 UTC - in response to Message 5152.
Last modified: 3 Jan 2009 | 5:03:26 UTC

thanks for the replies. MrS, now that I think about it, it probably was Seti wu's taking a dump while in a RDP session. Neil A., I would be using tightVNC, save for it doesn't work with XP64. I believe the latest pay VNC version does. Thanks for the link, I'll check it out.

ExtraTerrestrial Apes
Volunteer moderator
Volunteer tester
Avatar
Send message
Joined: 17 Aug 08
Posts: 2705
Credit: 1,311,122,549
RAC: 0
Level
Met
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 5212 - Posted: 3 Jan 2009 | 18:55:21 UTC

Didn't check the link, but from what you say Neil, I understand that you should be fine if you connect from a CUDA-capable machine (with a CUDA driver)? Or does it mean the terminal session always chooses some generic driver, which is not CUDA capable?

MrS
____________
Scanning for our furry friends since Jan 2002

Profile paul
Avatar
Send message
Joined: 16 Oct 08
Posts: 7
Credit: 3,999,754
RAC: 0
Level
Ala
Scientific publications
watwatwatwat
Message 6292 - Posted: 1 Feb 2009 | 12:10:28 UTC - in response to Message 5064.

MrS, Neil, etc:

NVidia driver 181.22 released about a week ago seems to have cured this problem. Our driver guy has confirmed the bug in previous nvidia drivers, has seen it cleared up with this latest version. I installed it alongside BOINC 6.5.0, might be a bit early to tell, but (knock on wood) I'm running RDP without issues now, even with seti gpu work.

ExtraTerrestrial Apes
Volunteer moderator
Volunteer tester
Avatar
Send message
Joined: 17 Aug 08
Posts: 2705
Credit: 1,311,122,549
RAC: 0
Level
Met
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 6455 - Posted: 5 Feb 2009 | 20:06:47 UTC - in response to Message 6292.

Thanks, it really works!

However, BOINC already had to be started when I logged on via remote desktop. If I (re)start BOINC remotely it still won't detect the CUDA device. Well, I can live with that.. now if only that stupid stock cooler on that 8600GT wouldn't get so loud while letting the chip reach 80°C.. :p

MrS
____________
Scanning for our furry friends since Jan 2002

ExtraTerrestrial Apes
Volunteer moderator
Volunteer tester
Avatar
Send message
Joined: 17 Aug 08
Posts: 2705
Credit: 1,311,122,549
RAC: 0
Level
Met
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 6483 - Posted: 7 Feb 2009 | 0:40:43 UTC - in response to Message 6455.

Actually.. now the WUs don't error out when I connect via remotedesktop. But after I do this once the current WU is finished and afterwards BOINC switches into cpu emulation mode (seti). GPU-Grid doesn't have emulation.. what's happening there? (don't want to test myself, as the card is too slow for GPU-grid)

MrS
____________
Scanning for our furry friends since Jan 2002

Post to thread

Message boards : Graphics cards (GPUs) : RDP and CUDA

//