This means you likely have one of two problems:
1) You have not updated your copy of AMBER 12 with all the latest bug
fixes. Check what version is reported for the GPU code in mdout. Should be
12.2. If it is not this then run patch_amber.py --update a few times until
it reports everything is updated then rebuild and try again.
2) You have problems with the system you are trying to simulate. Perhaps
it is not stable, bad parameters, bad structure etc. I would take a
careful look at the simulation, where it stops, what the structure looks
like immediately before it stops, if you can try with ntpr=1, ntwx=1. Try
running on the CPU code and see what happens etc. The error you are
showing typically happens when you have NaN's in the force array.
All the best
Ross
On 3/18/13 11:10 AM, "Albert" <mailmd2011.gmail.com> wrote:
>On 03/18/2013 05:17 PM, Ross Walker wrote:
>> Hi Albert,
>>
>> Use the serial version (pmemd.cuda) and see if that runs ok. GTX590's
>>and
>> 690's lack the PCI-E bandwidth to run successfully in parallel. With 2
>> GTX590s in your node I suggest running 4 independent jobs.
>>
>> All the best
>> Ross
>
>
>Hi Ross:
>
> thanks a lot for kind advices. It also failed, but with different
>messages:
>
>
>Error: unspecified launch failure launching kernel kReduceForces
>cudaFree GpuBuffer::Deallocate failed unspecified launch failure
>
>
>best
>Albert
>
>_______________________________________________
>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 Mon Mar 18 2013 - 16:00:04 PDT
