Re: [AMBER] pmemd.cuda errors on GTX400 series cards - USER POLL

From: Christophe Deprez <christophe.deprez.bri.nrc.ca>
Date: Thu, 09 Sep 2010 12:11:45 -0400

  Hi Sasha,

I can report very similar experience on our system:

SuperMicro SYS-7046GT-TRF, single Intel Xeon X5650, 12GB RAM,
3x *NVIDIA GTX 470* from MSI (N470GTX M2D12),
CentOS 5.2 x86_64, gfortran 4.1.2-42,
nvcc 3.1, V0.2.1221, NVIDIA Driver 256.40,
Amber11 with bugfixes 1 to 7 applied, compiled with './configure -cuda
gnu' (SPDP by default).

I've been running the JAC NPT simulation using the inpcrd, prmtop, mdin
files provided by Ross in his September 8 post.
The jobs are repeatedly stopping, respectively after 95,000 steps (196
ps), 421,000 steps (848 ps), 53,000 steps (112 ps), 43,000 steps (92 ps)
with the same error message: "Error: the launch timed out and was
terminated launching kernel kPMEGetGridWeights".

Christophe

On 09/05/2010 06:47 PM, Sasha Buzko wrote:
> Hi everyone,
> we are trying to figure out whether there is a hardware/driver problem
> with consumer Fermi cards (GTX 400 series). One observed error occurs
> almost randomly and produces the following output:
>
> Error: the launch timed out and was terminated launching kernel kPMEGetGridWeights
>
>
> If you have experienced this error on your hardware while running
> pmemd.cuda, please post a reply with the model of your video card, CUDA
> driver version and which pmemd.cuda mode was used (SPDP, SPSP or DPDP).
>
> If you have seen no errors on this hardware and produced reasonable
> results, please post as well.
>
> Thank you in advance
>
> Sasha
>
>
>
> _______________________________________________
> AMBER mailing list
> AMBER.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Thu Sep 09 2010 - 09:30:05 PDT
Custom Search