RE: problems running GPU-accelerated namd

From: Vermaas, Joshua (Joshua.Vermaas_at_nrel.gov)
Date: Mon Dec 10 2018 - 12:10:06 CST

The +devices argument isn't needed here, and should probably instead be looking at only the +p argument to specify the number of processors. NAMD's error message is saying that because only 1 video card (device) was found, it can't use the 6th one to run NAMD, so it errors out. My typical GPU call looks like this:

/PATH/TO/GPU/NAMD/namd2 +p8 input.namd > input.log

Only if you have multiple GPUs on the hardware and you want to only use a subset (maybe because you want to run two simulations on the same machine) do you need to specify +devices.

-Josh

On 2018-12-10 08:20:35-07:00 owner-namd-l_at_ks.uiuc.edu wrote:

Hello,

What is the output of 'nvidia-smi' ?

Best,

Victor
On Sun, Dec 9, 2018 at 3:03 AM 李耀 <liyao17_at_mails.tsinghua.edu.cn> wrote:
&gt;
&gt; Dear NAMD users,
&gt;
&gt; I'm running namd2 on GPU server with NAMD_Git-2017-12-25_Linux-x86_64-multicore-CUDA and this is the command:
&gt; ./namd2 +devices 5 ~/4ntw/string_eq_5ns/f1t10/win_f1/win5_f1.conf &gt; ~/4ntw/string_eq_5ns/f1t10/win_f1/win5_f1.log
&gt;
&gt; It came out the following error:
&gt; Charm++ fatal error:
&gt; FATAL ERROR: Pe 0 unable to bind to CUDA device 5 on omnisky because only 1 devices are present
&gt;
&gt; Aborted (core dumped)
&gt;
&gt; Then I declare 8 CPUs:
&gt; ./namd2 +p 8 +devices 5 ~/4ntw/string_eq_5ns/f1t10/win_f1/win5_f1.conf &gt; ~/4ntw/string_eq_5ns/f1t10/win_f1/win5_f1.log
&gt;
&gt; The coming message is:
&gt; Charm++ fatal error:
&gt; FATAL ERROR: Pe 4 unable to bind to CUDA device 5 on omnisky because only 1 devices are present
&gt;
&gt; Aborted (core dumped)
&gt;
&gt; I have a few questions:
&gt; 1) Do I need to declare the number of CPU to use(+p num) in this version of NAMD?
&gt; 2) How to make all the calculation(configuration and energy calculation) run with GPU but not with GPU and CPU?
&gt; 3) How to explain those errors? There seems no problem about the GPU devices.
&gt;
&gt; Thank you for reading my mail.
&gt;
&gt; Best,
&gt; Yao Li
&gt;

</liyao17_at_mails.tsinghua.edu.cn>

This archive was generated by hypermail 2.1.6 : Tue Sep 17 2019 - 23:20:23 CDT