From: sonia ziada (sonia.ziada_at_univ-orleans.fr)
Date: Thu Nov 17 2016 - 01:54:59 CST
Hello,
Did you check the % of utilization of your GPU card with the commands
"nvidia-smi -q -g 2 -d UTILIZATION -l" or "nvidia-smi dmon -s u" ?
Check also the gpu card driver, did you install the right one ?
If you are using several GPU cards, you should specify also the argument
devices, as follow:
namd2 +idlepoll +p<nbprocess> +devices <num of gpu card, ...> <config
file>. Personally, I do not use the command charmrun, so try without.
Best Regards.
Sonia Ziada, ph.D. student
ICOA laboratory, Structural Biology and Chemoinformatics team.
Orléans, France.
On 11/17/2016 08:27 AM, Atanu Maity wrote:
> Dear experts,
>
> We are not getting expected benchmark for the acceleration of
> NAMD2.11 simulation speed using Multiple CPUs and GPU.
>
> The configuration of the CPU processor is "Intel(R) Xeon(R) CPU
> E5-2650 v3 @ 2.30GHz" and we are using NVIDIA Tesla K40c GPU card.
>
> The benchmark time of the simulation of a system containing ~1,70,000
> atom is as follows
>
>
> No. of CPU ns/day (CPU) ns/day (CPU+K40c)
> 32 2.244 2.087
> 16 2.760 3.002
> 8 2.591 2.582
> 4 1.984 2.051
> 2 1.141 1.145
> 1 0.591 0.590
>
>
> The NAMD version is 'NAMD_2.11_Linux-x86_64-multicore-CUDA' and the
> command line is as follows -
> charmrun +p32 namd2 +idlepoll input.namd > output.log
>
> Can anyone please explain or suggest the probable reasons why we are
> not getting the acceleration using GPU ?
>
>
> With regards
> Atanu Maity
> Bioinformatics Center
> Bose Institute
>
>
This archive was generated by hypermail 2.1.6 : Tue Dec 27 2016 - 23:22:37 CST