From: Francesco Pietra (chiendarret_at_gmail.com)
Date: Tue Aug 02 2011 - 16:52:01 CDT

Dear John:
I rely on the cuda driver installed through Debian amd64, and with
version 275.09.07-5 NAMD runs fine with two GTX-470 cards, at standard
temperatures for such a card (I don't call the X server in this case).
So, there was no burned card.

With VMD, at some intermediate version, there was partial success with
VMD; with this version of the driver, launching VMD hangs the
X-server, and I have to reboot the GPU machine through a ssh-linked
vintage desktop (which is slow with VMD, but it works).

I am saying that because I came across your mail below that I had
forgotten. I'll look for the possibility to install the 270 version
(is that a legacy driver?) through Debian, although I have also to run
NAMD, and this presently runs fine. In addition, I'change in a few
days the two GTX-470 with two GTX-580. Is the nvidia driver the same
for both?

********
On Tue, Jul 5, 2011 at 3:01 PM, John Stone <johns_at_ks.uiuc.edu> wrote:
>
> Francesco,
> If you're having trouble with the 275 series driver, I would
> recommend installing the driver version that was posted with CUDA 4.0,
> which is version 270.41.19. We are using 270.41.19 on all of the machines
> here and we don't have any problems currently.

have no problems with that driver version with both a GeForce GTX 480
and a GeForce GTX 560 Ti here.

[akohlmey_at_fermi Downloads]$ cat /proc/driver/nvidia/version
NVRM version: NVIDIA UNIX x86_64 Kernel Module 275.09.07 Wed Jun 8
14:16:46 PDT 2011
GCC version: gcc version 4.4.4 20100630 (Red Hat 4.4.4-10) (GCC)

based on francesco's previous reports,
i would suspect broken hardware.

i would also recommend to check for overheating.

i've recently experienced a case where an overheated
CPU (due to throttling) would drag a machine to the
speed of geological processes which made it look
as if it was crashed.

cheers,
    axel.

> Cheers,
> John Stone
***********

thanks
francesco pietra