Re: [AMBER] "cudaDeviceEnablePeerAccess all CUDA-capable devices are busy or unavailable"

From: Ross Walker <ross.rosswalker.co.uk>
Date: Thu, 08 May 2014 01:59:05 -0700

Yes - to run AMBER 14 pmemd.cuda.MPI you MUST set the GPUs to default
compute mode. As root:

nvidia-smi -c 0

I wish NVIDIA (or Linux) would allow this to be set in user space but alas
not. One could setuid root for the nvidia-smi binary I guess but that
would only work if you implicitly trust all your users.

All the best
Ross



On 5/8/14, 1:31 AM, "ahmed.sajid.stfc.ac.uk" <ahmed.sajid.stfc.ac.uk>
wrote:

>Hi,
>
>Sorry again for ignoring the real question. I'm trying to do Multi-GPU
>run by setting DO_PARALLEL. I guess my coffee hasn't kicked in yet.
>
>Regards,
>Ahmed.
>--
>Scanned by iCritical.
>
>_______________________________________________
>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 May 08 2014 - 02:30:03 PDT
Custom Search