From: Giacomo Fiorin (giacomo.fiorin_at_gmail.com)
Date: Thu Jan 28 2021 - 19:01:54 CST
Hi Raghav, without knowing the details of how NAMD and GROMACS were built
on your cluster, advice can only be quite generic. There are a lot of
factors that may affect how each code goes.
Two general pieces of advice:
1) Review each line of input in the NAMD and GROMACS scripts, and make sure
you are using the same parameters in each (cutoffs, pairlist frequencies,
etc), and that GROMACS is built with double precision (unless you're
running both codes on GPU nodes). You need to compare apples with apples.
Speaking from experience (I built and benchmarked both codes for my
colleagues on our internal cluster) I found GROMACS to be faster at lower
node counts, and NAMD overtaking it at higher node counts before scaling
out. This with equivalent inputs, though.
2) Pick the code and build configuration that work best for the given set
of features and the problem that you are dealing with. It is common to use
multiple
codes on the same project, when each has unique advantages or features.
And there are also other codes that can handle well a system your size with
CHARMM force field on different hardware including single-GPUs.
Giacomo
On Mon, Jan 25, 2021 at 7:05 AM raghav singh <raghavbioinfo_at_gmail.com>
wrote:
> Dear NAMD Developers and Users,
> I have a protein(solvated and neutralized with 0.15mM NaCl concentration)
> simulation system under classical MD scenario with charmm36 ff parameters
> implemented in VMD/NAMD.
> I have prepared my system with both GROMACS and NAMD* (*preferred because
> I would like to run Accelerated MD later on). The system consists of 148943
> atoms all together, I have tried to test the speed and scaling of NAMD
> using 1,2,4,8,16 and 32 nodes and found a linear scaling. I get ~50ns/day
> of yield with NAMD on 32 nodes (768 processing cores), while if I run the
> similar system with gromacs on 20 nodes (480 processing cores) and I get
> ~270ns/day.
>
> Would it be possible for someone in the group to look at my issue, as
> ~150000 atoms simulation system should not require this much of resources
> to run this short simulation. I think I am wasting my HPC resources and I
> am in need of dire intervention from experts.
>
> Please find Attached the configuration and run script. I would greatly
> appreciate your expert intervention and help, so that I can make the
> best use of the HPC resources.
>
> Thanking you
>
> Best regards
> Raghav
>
This archive was generated by hypermail 2.1.6 : Fri Dec 31 2021 - 23:17:10 CST