[AMBER] Amber 20 - Support Multi GPU

From: Adriano José Ferruzzi <adrianof.unicamp.br>
Date: Wed, 22 Jul 2020 16:31:29 -0300

Hello,

We recently purchased the license for Amber 20 and compiled it with the
following configuration:

GCC 4.8.5
OpenMPI 3.1.3
CUDA 10.1
Cmake 3.15

And these flags on run_cmake:

cmake $AMBER_PREFIX/amber20_src \
    -DCMAKE_INSTALL_PREFIX=$AMBER_PREFIX/amber20 \
    -DCOMPILER=GNU \
    -DMPI=TRUE -DCUDA=TRUE -DINSTALL_TESTS=TRUE \
    -DDOWNLOAD_MINICONDA=TRUE -DMINICONDA_USE_PY3=TRUE \
    2>&1 | tee cmake.log

We followed Amber20 installation guide presented in this manual
https://ambermd.org/doc12/Amber20.pdf

We are now trying to run pmemd.cuda_SPFP.MPI using multi GPU to take
advantage of the 2 GPUs Tesla k40 we have per node, but Amber starts to run
and gets 0% on both GPUs. Here, we followed the instructions presented at
https://ambermd.org/gpus16/index.htm. Unfortunately, we didn't have success.

The same job works perfectly when using the single GPU (with
pmemd.cuda_SPFP executable).

We would appreciate your help on this.

Thank you in advance,
Cheers.

-- 
Adriano Ferruzzi
Systems Administrator
Institute of Chemistry - Unicamp
CCES - Center for Computing in Engineering & Sciences
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Wed Jul 22 2020 - 13:00:04 PDT
Custom Search