P.S. I've already tried to set up manually GPUs for each input script like
# script 1
export CUDA_VISIBLE_DEVICES="0"
# script 2
export CUDA_VISIBLE_DEVICES="1"
is there any possibility to define it in some automatic way e.g.
always take GPU which is not being used ? or alternatively divide gpus
according to the number of the runned scripts (jobs )?
Il giorno gio 4 ago 2022 alle ore 12:10 Enrico Martinez
<jmsstarlight.gmail.com> ha scritto:
>
> I've just been able to install amber with CUDA on my 2 gpus work-station.
>
> In order to run job I use the following script :
>
> #!/bin/bash
> pmemd.cuda -O -i ./in/md_production_330K.in -o prod.out -p
> protein.prmtop -c equil1p.rst -ref equil1p.rst -r prod.rst -x
> prod.netcdf
>
>
> 1) What should I add to define which GPU will be used?
>
> I checked nvidia-smi and found that pmemd uses the gpu 0
>
> | 0 N/A N/A 2162 G /usr/lib/xorg/Xorg 4MiB |
> | 0 N/A N/A 2783 G /usr/lib/xorg/Xorg 4MiB |
> | 0 N/A N/A 6465 C pmemd.cuda 913MiB |
> | 1 N/A N/A 2162 G /usr/lib/xorg/Xorg 110MiB |
> | 1 N/A N/A 2783 G /usr/lib/xorg/Xorg 213MiB |
> | 1 N/A N/A 3656 G /usr/bin/gnome-shell 163MiB |
> | 1 N/A N/A 5985 G /usr/bin/nvidia-settings 0MiB |
> | 1 N/A N/A 6063 G ...RendererForSitePerProcess 63MiB |
> | 1 N/A N/A 6720 G ...692524829162169360,131072 97MiB
>
> However, in the nvidia X server I see that the GPU 1 is being heated
> (86 C), although the fan of GPU 0 (55 C) is working very extensively.
>
> 2) May I run several (e.g. 10 jobs) using 2 gpus (5 jobs on each gpu?)
> If so how should I parametrize the script?
> Many thanks in advance
> Cheers
> Enrico
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Thu Aug 04 2022 - 13:38:06 PDT
