Take the 2-minute tour ×
Server Fault is a question and answer site for professional system and network administrators. It's 100% free, no registration required.

I recently installed a few dual-GPU CUDA workstations (two dual gtx480s and a dual gtx470) with no apparent trouble. I just tried a gtx480/ Tesla C2050 and not only does deviceQuery fail with a weird error - something about an invalid device ID or count or something - but now the machine won't boot into graphical mode (remote access via ssh still works fine). I assume I am doing something tragically wrong... Is mixing these GPUs even possible? If so, does it require special configuration? I must have missed something basic. Any help is appreciated - I am new to this and need all the help I can get.

EDIT:

Sorry, I thought I mentioned the OS. It's 64-bit CentOS 6, based on RHEL 6. The error is "invalid device ordinal" or some such. Imm going to try a few things and get back with more info, but I appreciate the suggestions so far, and will try migrating to superuser or unix... Thanks!

SOLVED:

So, in desperation, I tried reseating the Tesla C2050. Guess what? I only had one of the two power connectors in (!). Weirdest thing was that although it showed up fine with lspci and every other way I could check it, it was causing X to fail and deviceQuery to crap out. I guess the moral of the story is that I need to be less of a muppet. If anybody else is having the same problem... something to check!

share|improve this question
    
You could start by giving us more details, i.e. what OS are you using? Also, I suspect this question is a better fit for unix.se or SuperUser. –  Sven Sep 29 '11 at 12:58
    
@SvenW This question is regarding professional 3D workstations, which does fit in our FAQ (even though we lack much depth to handle these). –  sysadmin1138 Sep 29 '11 at 13:06
    
I doubt any of the SE sites are going to know much about mixing Tesla cards with normal graphics cards. The only place I've seen Teslas they were loaded up in headless "compute" servers. As much as they cost the reseller or nVidia should be all over supporting any "minor" problems your having. –  Chris S Sep 29 '11 at 13:10
    
I'll agree and was only thinking that unix.se in particular might have more knowledge on how to get X running in such a special case. –  Sven Sep 29 '11 at 13:29

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

Browse other questions tagged or ask your own question.