Re: [AMBER] Info About GPU and PMEMD versions

From: Jason Swails <>
Date: Wed, 22 Oct 2014 08:55:33 -0400

On Wed, 2014-10-22 at 10:55 +0200, Lorenzo Gontrani wrote:
> Dear Amber experts,
> we have acquired/refurbished/set-up a computing cluster composed of
> several different nodes, some of which have GPUs, that I have tried to
> use with Amber12 but unsuccessfully.
> The graphic cards are:
> nVIDIA Fermi M2050

This works with Amber 12.

> nVIDIA Tesla M1070

This works with Amber 11. I believe one of the Amber 12 updates dropped
support for the Tesla architecture.


This should work with Amber 12.

> The support for these cards is different among the various version of
> PMEMD, (we have access to a copy of Amber12 and one of Amber11).
> Could anyone give me some advice about which PMEMD version should be run
> on these GPU (if possible)? Is the information about old PMEMD versions
> still available on the website?


> It is a great pity to stick to simple MPI..GPU is another story.

The M1070 may be worth abandoning for Amber. It is not well supported
at all, and getting Amber 11 and Amber 12 both running may be
challenging (I'm not sure if there is one version of the CUDA toolkit
that is compatible with both of them).


Jason M. Swails
Rutgers University
Postdoctoral Researcher
AMBER mailing list
Received on Wed Oct 22 2014 - 06:00:06 PDT
Custom Search