Hello,
See the amber website describing GPU usage: http://ambermd.org/gpus/
There is a section that says "Supported GPUs". In that section reads this
warning:
There are currently known issues with GTX465, GTX470, GTX480 and GTX580
cards hanging after unspecified periods of time. We are working with NVIDIA
to try to find the cause of this so that we can fix / workaround it.
There has been a lot of discussion on this list about this very topic, with
the end result being the above disclaimer. As of yet, there is no fix to
the best of my knowledge.
Good luck!
Jason
On Fri, Dec 10, 2010 at 8:14 AM, Gabriel Urquiza <urquizagabes.gmail.com>wrote:
> Greetings amberists,
>
> I am having some problems running PMEMD simulations of a protein in a water
> box. The system has around 60.000 atoms and I'm running it CUDA on a
> GeForce
> GTX 480 1.4Ghz board with a global memory size of 1535 MB. I am running
> several short dynamics with this system and it turns out that when I check
> at the human-readable output files I notice that some of them just
> abnormally terminated. The step in which the simulations finish doesn't
> follow any apparent pattern, some terminate near the end, others right at
> the beginning. Most of the simulations runs through though but if I remove
> every file, restart the machine and resubmit all the simulations, using the
> same script, in the same order, it is almost certain that the simulations
> that crashed earlier won't be exactly the ones that will crash now. It
> seems
> a very random thing and I find it very weird. I am placing my bets on the
> narrow memory size of the board I'm using, but I've decided to ask for some
> light on the matter.
>
> The machine is running on double precision and it has an i7 processor. The
> nohup pops one out of two error messages down below, whenever a dynamic
> crashes:
>
> Error: the launch timed out and was terminated launching kernel
> kCalculatePMENonbondForces
> Error: the launch timed out and was terminated launching kernel
> kPMEGetGridWeights
>
> I hope you can assist me with this.
>
>
> Gabriel Urquiza
>
> ----------------
>
> Universidade Federal da Paraíba - UFPB
> Departamento de Química - Programa de Pós-Graduação
> _______________________________________________
> AMBER mailing list
> AMBER.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber
>
--
Jason M. Swails
Quantum Theory Project,
University of Florida
Ph.D. Graduate Student
352-392-4032
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Fri Dec 10 2010 - 08:00:03 PST
