Hi Ashutosh,
We'll need some more info in order to reproduce the problem you are seeing
- firstly can you confirm that this works fine with a single GPU - with
AMBER 12 you really don't gain much by using multiple GPUs - You are much
better off running multiple calculations, one on each GPU, to get better
statistics. As such we haven't focused on extensive debugging of the MPI
GPU code, especially since it will change radically with the next release
which we are working on right now.
So it is possible there is a bug in the code, in the mpi implementation or
something just weird with the system you are trying to simulate. If you
can let us know which compiler you are using, which MPI, which cuda and
driver version, which GPU types and provide a copy of your input files we
can try to investigate further.
All the best
Ross
On 11/19/13 7:47 AM, "Ashutosh Shandilya" <izerokelvin.gmail.com> wrote:
>I think extra point issue is resolved but this
>(gpu_dowload_partial_forces:download_failed unspecified launch failure) is
>still there. As suggested
>I checked it gave the the same output.
>
>Attached is the screen shot. Please help.
>
>
>On 18 November 2013 23:02, Daniel Roe <daniel.r.roe.gmail.com> wrote:
>
>> Hi,
>>
>> On Sun, Nov 17, 2013 at 10:41 PM, Ashutosh Shandilya
>> <izerokelvin.gmail.com> wrote:
>> > Earlier I got different errors which were sorted out
>> >
>> > numextra > 0 (in parallel) *Extra points are only supported in the
>>serial
>> > GPU code at present.*
>> > I made all the required changes without any extra point etc when it
>> should
>> > run but I still got this error.
>>
>> It would help if you were more specific about what the system you are
>> trying to simulate is (are you e.g. trying to use the TIP4P water
>> model?) and what specific steps you've taken to try and solve the
>> problem. If your topology has any extra points it will not run using
>> pmemd.cuda.MPI. If you're not supposed to have any extra points in
>> your topology, it's possible something strange is going on. If you
>> would like, send me your topology (off-list) and I will look it over.
>>
>> > I looked at internet it says to include bugfixes. I have installed all
>> the
>> > bugfix still the problem persists.
>>
>> Make sure you see this in your output:
>>
>> |--------------------- INFORMATION ----------------------
>> | GPU (CUDA) Version of PMEMD in use: NVIDIA GPU IN USE.
>> | Version 12.3.1
>>
>> If the version number is any lower you don't have the most recent
>>bugfixes.
>>
>> -Dan
>>
>> >
>> > Please have a look
>> > _______________________________________________
>> > AMBER mailing list
>> > AMBER.ambermd.org
>> > http://lists.ambermd.org/mailman/listinfo/amber
>>
>>
>>
>> --
>> -------------------------
>> Daniel R. Roe, PhD
>> Department of Medicinal Chemistry
>> University of Utah
>> 30 South 2000 East, Room 201
>> Salt Lake City, UT 84112-5820
>> http://home.chpc.utah.edu/~cheatham/
>> (801) 587-9652
>> (801) 585-6208 (Fax)
>>
>> _______________________________________________
>> 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
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Tue Nov 19 2013 - 08:30:04 PST
