Are these timings fine?

User discussions
, , , ,
1

GROMACS version: gromacs/2021.4_mpi_plumed
GROMACS modification: No

Hello,
I am running Gromacs on HPC for the first time, and I am unsure if the timings are fine. I am not sure whether I have allocated too many or too few resources.

The system consists of 57,810 atoms.
The timings are listed below:

NVT

          Core t (s)   Wall t (s)        (%)
Time:    22796.285      474.923     4800.0
                   (ns/day)    (hour/ns)
Performance:       18.193        1.319

NPT

               Core t (s)   Wall t (s)        (%)
       Time:    23881.127      497.524     4800.0
                 (ns/day)    (hour/ns)
Performance:       17.366        1.382

Currently running production from 0 to 10, and then will start 10 to 100.

As per GROMACS description…
“Running on 1 node with total 48 cores, 48 logical cores.[…] Brand: Intel(R) Xeon(R) Platinum 8268 CPU @ 2.90GHz”

What do you think about those timings? Once I have observed that ns/day can reach over 30-40ns/day, so it appears that my results are rather subpar…
Would running it on a GPU be a wise decision? I have access to those, but only nine processors are available. As a result, they are often occupied.

Thank you in advance

2

Dear lutek14,

Performances are in general very system and architecture dependent. You should benchmark your machine to get an idea of the maximum you can get out of it.

Most of the times GPUs increase the performances of the simulations. But again, you should check which combination of GPUs/CPUs gets the best performance on your set up and with your system.

3

Hell obZehn,

Could you suggest any recommended methods to conduct this benchmark?

4

Yes, run simulations! Build a short MD sim of a few hundreds of ps and test it with different number of threads etc. Then, compare the ns/day and pick your best choice. You can take a look at gmx mdrun --h to see which flags you have available, but the usage of them will again depend on the GROMACS built you have (GPU? MPI/thread MPI?) and where specifically you are running the simulation.

If you are not familiar with some of these words and/or these optimization aspects, tale a look at this relevant page. Hope this helps!