On Mon, Sep 13, 2021, ARITRA MITRA wrote:
>I am facing issues while installing Amber18 (with AmberTools 19) in my
>HPE Proliant DL380 Gen10 Server (with 2*NVidia Tesla A100 GPU Cards). The
>serial version installs without any problem. But while building the
>parallel version using:
>./configure -cuda gnu
The above is a bit confusing: to get the parallel cuda version
(pmemd.cuda.MPI) you need to give both the -mpi and -cuda flags to
configure.
>CentOS Linux release 7.9.2009 and
>Cuda compilation tools, release 10.2, V10.2.89.
It is certainly hidden, but if you look at the Amber web site, and go to
GPU Support -> Hardware, you'll find this:
"Ampere (SM_80) based cards require CUDA 11.0 or later." It's hard to trap
this problem at compile time, since we have no waying of knowing then what
hardware will eventually be used. Still, at a minimum, you will need vesion
11 of the CUDA SDK.
[Amber CUDA developers: is there a way to emit a useful error message if
someone tries to run pmemd.cuda on unsupported hardware?]
Finally, I can't easily tell if Amber18 supports the SM_80 architecture or not.
Maybe others on the list will know the answer here.
....regards....dac
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Mon Sep 13 2021 - 12:30:02 PDT
