Re: [AMBER] CUDA problems

From: Ross Walker <>
Date: Mon, 23 Mar 2015 09:52:29 -0700

Hi John,

What operating system are you using? and how were the NVIDIA drivers installed in the first place?

If it is redhat and they are through an RPM you can do - yum list | grep nvidia - and then uninstall all the installed RPMs that are related to NVIDIA and then reinstall them.

All the best

> On Mar 23, 2015, at 7:41 AM, Beale, John <> wrote:
> CUDA was installed on my Linux workstation (which possesses two Nvidia GPUs) and it worked perfectly with Amber. Unfortunately, I mistakenly installed several recommended software updates, a number of which were for the Nvidia units, and now I can't get CUDA to work. Initially, the deviceQuery and bandwidthTest tests passed. Now they fail. The deviceQuery gives the following response:
> Warning: Module nvidia_uvm_346.46 not found
> cudaGetDeviceCount returned 30
> ? Unknown error
> Result = FAIL
> In my /var/lib/dkms directory I have the nvidia-uvm directory (note that this name is hyphenated, not underscored as it is in the deviceQuery error message). In /usr/src I have an nvidia-uvm-346.46 directory.
> If someone could help me get past these problems and get my CUDA working again I would appreciate it greatly! I have found on the web that others have had this problem, but I haven't found a solution there.
> John M. Beale, Jr. John M. Beale, Jr., Ph.D.
> Professor of Medicinal Chemistry and Pharmacognosy
> Saint Louis College of Pharmacy
> 4588 Parkview Place
> Saint Louis, Missouri 63110-1088
> Office: 314-446-8461
> Cell: 314-315-0409
> FAX: 314-446-8460
> _______________________________________________
> AMBER mailing list

AMBER mailing list
Received on Mon Mar 23 2015 - 10:00:05 PDT
Custom Search