Re: [AMBER] Testing pmemd.cuda.MPI - how many failed file comparisons are innocuous?

From: Nikolay N. Kuzmich <nnkuzmich.gmail.com>
Date: Fri, 12 Feb 2021 19:57:35 +0300

Dear David,

*> *
*>I have installed the GPU version of pmemd (Amber 20) in parallel and ran *
*>the tests for both 2 and 4 GPUs. The numbers of failed file comparisons *
*>were 13 out of 135 and 33 out of 189 for 2 and 4 GPUs, respectively.
These *
*>results worry me a bit... *

Can you say exactly what test you ran, i.e. what command did you type to
run
the tests?

...thanks....dac

that's what I did for the 2 cards right after make install:

source $AMBERHOME/amber.sh
export LD_LIBRARY_PATH=/usr/local/cuda-11.2/lib64
export DO_PARALLEL="mpirun -np 2"
sudo nvidia-smi -c 0
export CUDA_VISIBLE_DEVICES=0,1 *[I am not actually sure if it was
necessary - N.K.]*
make test

and after that for the 4 cards:
export DO_PARALLEL="mpirun -np 4"
make test

Kind regards
Nick
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Fri Feb 12 2021 - 09:00:02 PST
Custom Search