Fwd: SOLVED: about latest nvidia cuda driver 275.09.07

[Francesco Pietra <chiendarret@gmail.com>]

Also molecular dynamics is running fine with the new driver by simply
activating the two gtx470 cards with

#  nvidia-smi -L

#  nvidia-smi -pm 1

the latter for 'persistent mode'

Great surprise: with the new driver - on the same system of 250,000
atoms - it is running notably faster than before.

fp


---------- Forwarded message ----------
From: Francesco Pietra <chiendarret@gmail.com>
Date: Wed, Jul 20, 2011 at 7:47 AM
Subject: SOLVED: about latest nvidia cuda driver 275.09.07
To: amd64 Debian <debian-amd64@lists.debian.org>, Lennart Sorensen
<lsorense@csclub.uwaterloo.ca>


With the new nvidia-kernel-dkms 275.09.07-5 in Debian GNU-Linux amd64
wheezy, all described problems with the CUDA-accelerated viewer VMD
vanished. Thus, my suspicion that problems (absent with driver
270.41.19-1) were introduced by driver 275.09.07-1 was correct.

I am not at molecular dynamics in these days, but all prospects are
that CUDA problems are solved for MD too by the new driver.

francesco pietra


---------- Forwarded message ----------
From: Francesco Pietra <chiendarret@gmail.com>
Date: Wed, Jul 13, 2011 at 3:57 PM
Subject: Fwd: about latest nvidia cuda driver 275.09.07
To: amd64 Debian <debian-amd64@lists.debian.org>, Lennart Sorensen
<lsorense@csclub.uwaterloo.ca>


As to my problems with this cuda computer, which started with
upgrading to the nvidia driver 275.09.01-1, other people have my same
feeling that there is no hardware problem (someone had raised doubts
that a crossfire can't support more than one nvidia card, while SLI is
not needed for cuda), possibly a driver problem.

My question is (please, see below) how to try to resolve the puzzle,
by either taking the latest driver from "unstable" or back to the
older 270.41.19-1 driver. I have no experience in doing either move
safely.

Thanks for advice

francesco pietra

 Originally Posted by chiendarret  View Post
Also, before current nvidia driver 275.09.01-1, I worked for a week
with 270.41.19-1 without any warning.
That makes me think its just an issue with the 275.09.01-1 driver, not
your hardware. You could try getting the drivers from 'sid' (currently
275.09.07-5), or maybe even experimental if you were game. Or you
could manually install 270.41.19 drivers. There might even be aq way
to rollback to the 270.41.19-1 drivers as well, but I dont know
how.....if its even possible.


---------- Forwarded message ----------
From: Francesco Pietra <chiendarret@gmail.com>
Date: Sat, Jul 9, 2011 at 12:04 AM
Subject: Fwd: about latest nvidia cuda driver 275.09.07
To: amd64 Debian <debian-amd64@lists.debian.org>


Under certain circumstances, especially if the CUDA machine is
shutdown or reboot by commands from a ssh-linked desktop, problems
vanish. I begin to suspect that the combination of mutithreading with
the two gtx-470 cards may find problems with the "composite" extension
of X11. When VMD (an OpenGL viewer CUDA accelerated) is launched from
a gnome terminal it says:

Multithreading, 6 processors

Available two CUDA accelerations

Detected X11 "Composite" extension:if incorrect display occurs, try
disabling this optional X server feature

OpenGL renderer GeForce gtx 470/PCI/SSE2

Full GSL rendering mode available



In this case there were no problems in using VMD with a 8GB file,
using many of its features (removing water, calculating RMSD, making
movie of the MD trajectory, etc. What happens - i.e, what is left
available - if "composite" extension is disabled I have not tried.

francesco




---------- Forwarded message ----------
From: Francesco Pietra <chiendarret@gmail.com>
Date: Thu, Jun 30, 2011 at 10:21 AM
Subject: about latest nvidia cuda driver 275.09.07
To: amd64 Debian <debian-amd64@lists.debian.org>


Following kind guidance by Lennart Sorensen as to the correct
installation of the nvidia driver, I recently got a two GTX-470
computer work perfectly for molecular dynamics simulations with
NAMD-CUDA.

The simulations were launched from linux terminal, without calling the
X server. Command (as root)

nvidia-smi -L

was first needed to activate the GTX 470 cards (udev normal affair)
********************

On 27Jun the system was upgraded from driver 270.41.19 to 275.09.07.
On rebooting, dkms did its job for the existing headers 2.6.38-2.

However, now the above NAMD-CUDA MD launching does not work any more.
Sincerely I can't say if the crash described below occurred
immediately after the driver upgrading, but the dates are very close.
Now, on launching MD as above described, the system hangs, the screen
shows blinking characters and letters, and the power must be shut
down. The log of the simulation says

CUDA error cudaStreamCreate on P2 0

which is a normal message of NAMD when the graphic boards are not
seen. However, the computer should not hang.

Launching NAMD-CUDA from a terminal window of gnome-2, the procedure
runs correctly.

I understand that this report fails to analyze correctly what happens.
It is a simple warning to see if other amd64 users had problems with
the new nvidia driver.

thanks

francesco pietra