From: Vermaas, Josh (vermaasj_at_msu.edu)
Date: Fri Sep 09 2022 - 09:51:34 CDT
Are you using the CUDASOAIntegrate codepath? Otherwise, NAMD 3 will fall back to the NAMD 2 code path, which is why the performance is identical. See the NVIDIA blogpost about what you need to change in your configuration files. https://urldefense.com/v3/__https://developer.nvidia.com/blog/delivering-up-to-9x-throughput-with-namd-v3-and-a100-gpu/__;!!DZ3fjg!-DCHHcZKZJCPXqopcn4hN0TIEIBGEcM_CvfhOZlB6VllexQwRW4ZW7Y_wjeAhZ4_ZiF-YvOBNQyf5RxOI18$
-Josh
From: <owner-namd-l_at_ks.uiuc.edu> on behalf of SHIVAM TIWARI <t.shivam_at_iitg.ac.in>
Reply-To: "namd-l_at_ks.uiuc.edu" <namd-l_at_ks.uiuc.edu>, SHIVAM TIWARI <t.shivam_at_iitg.ac.in>
Date: Friday, September 9, 2022 at 4:39 AM
To: namd-l <namd-l_at_ks.uiuc.edu>
Subject: namd-l: No performance improvement with NAMD 3.0 alpha8 over namd 2.14 cuda
Dear all,
I am comparing the performances of namd 3.0 (alpha 8) and namd 2.14 on a system of 257k atoms with nvidia tesla k40 as my GPU. However, I don't see any performance improvement with namd 3.0 and both codes(3 & 2.14) are giving a benchmark of around 4 ns/day. How to get the best performance out of the available computational resources? Following is the SLURM script I am using for NAMD 3.0.
#!/bin/sh
#BATCH -A
#SBATCH --nodes 1
#SBATCH --ntasks-per-node=4
#SBATCH --account=project
#SBATCH --gres=gpu:1
#SBATCH --qos=project
#SBATCH --partition=high
#SBATCH --mem=10gb
#SBATCH --error=job.%J.err
#SBATCH --output=job.%J.out
#SBATCH --time=10-00:00:00
#SBATCH --no-requeue
module purge
#PATH=$PATH:/apps/precompiled/namd/NAMD_2.14_Linux-x86_64-multicore-CUDA
PATH=$PATH:/scratch/proj/nsmswamireddy/NAMD_3.0alpha8_Linux-x86_64-multicore-CUDA
##charmexec=../NAMD_3.0alpha12_Linux-x86_64-multicore-CUDA/charmrun
##namdexec=../NAMD_3.0alpha12_Linux-x86_64-multicore-CUDA/namd3
charmrun namd3 +p $SLURM_NTASKS +setcpuaffinity +idlepoll prod.conf > pasr21.log
regards
shivam
This archive was generated by hypermail 2.1.6 : Tue Dec 13 2022 - 14:32:44 CST