Hi Mayank,
> I would like to know whether it is possible to run AMBER on a cluster
> where
> each node has one GPU. If yes, what are the things that I have to
> differently?
It depends on what you want to actually run. Right now you can run
individual jobs on individual nodes using 1 GPU per node. You would just
need to craft your scripts correctly to submit these jobs to the queue.
If you mean can you run a single job in parallel across multiple nodes in a
cluster the answer is, right now, no. However, I plan to release a patch,
within a month, against AMBER 11 that will allow you to use multiple GPUs
for a single run. These can be within a single node or across multiple nodes
and it will use MPI v2.0. Speedup across 4 to 8 GPUs is anticipated
depending on the system size and job parameters.
Note you will need a VERY good interconnect between nodes. The minimum is
probably QDR Infiniband in PCI-E 8x slots, better would be QDR IB or dual
rail QDR IB in PCI-E 16x slots. With DDR or SDR IB the scaling will probably
be limited to 2 or 3 nodes at most although I don't have definitive numbers
right now.
So, stay tuned.
All the best
Ross
/\
\/
|\oss Walker
---------------------------------------------------------
| Assistant Research Professor |
| San Diego Supercomputer Center |
| Adjunct Assistant Professor |
| Dept. of Chemistry and Biochemistry |
| University of California San Diego |
| NVIDIA Fellow |
|
http://www.rosswalker.co.uk |
http://www.wmd-lab.org/ |
| Tel: +1 858 822 0854 | EMail:- ross.rosswalker.co.uk |
---------------------------------------------------------
Note: Electronic Mail is not secure, has no guarantee of delivery, may not
be read every day, and should not be used for urgent or sensitive issues.
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Tue Aug 24 2010 - 11:00:02 PDT
