I am trying to use VMD/NAMD to carry out a molecular dynamics simulation, and while running the simulation, task manager shows that only 30% of my GPU is being utilized. When VMD starts up, it only shows 4GB of VRAM detected, while my NVIDIA RTX 3080 has 10GB of VRAM. I installed the CUDA accelerated versions of VMD and NAMD, what do I do to fully use my GPU for MD simulations? My CPU is not bottlenecking the simulation, it has 12 cores/24 threads and is only at 5-10% utilization.
You might wanna take a look at this:
http://www.ks.uiuc.edu/Research/namd/2.12/ug/node90.html
In NAMD, the bonded, non-bonded, PME, implicit solvent terms are accelerated by GPU, whereas other terms integration, rigid bonds, grid force, collective variables are still calculated on CPU (These are disabled for GPU acceleration in NAMD). Therefore the GPU utilization is only 10%.
Try Following:
Add following to config file
CUDASOAintegrate on
Command to run:
namd3 +p1 +setcpuaffinity +idlepoll +devices 0 configfile.namd
Further Reading: https://www.pugetsystems.com/labs/hpc/molecular-dynamics-benchmarks-gpu-roundup-gromacs-namd2-namd-3alpha-on-12-gpus-2330/
- Badges
- Science topic
- Similar topics
- Instrumentation Engineering
- Instruments