Hi, we have this benchmark report of VASP using Nvidia GPUs V100 and A100 https://news.developer.nvidia.com/nvidia-gpu-accelerated-vasp-6-uses-openacc-to-deliver-15x-more-performance/#respond but it is lacking information. Do someone has some information about the authors or some detail about it? It seems impossible to reproduce their results.
Not my area of expertise. Are you running with VASP.6? The linked post refers to VASP.6 performance, but when I visit the VASP website, it tells me that VASP.6 has not been released yet (https://www.vasp.at/wiki/index.php/Category:VASP6). Which would be mean that nobody in the general public, i.e. outside the VASP development team, would be able to reproduce the posted numbers at this time. Have you tried asking for clarification in the VASP forums? That seems to be the primary support channel for VASP.
You are right, but they offer early access to “beta” versions to some users, so I have access. I also ask information on VASP forum but I know that Nvidia guys follow that development. It is possible that the authors of this work are around here. Or someone else that also did that test. You know, small world.
When you say “impossible to reproduce the posted numbers”, how far off are those numbers from those in the linked post? If VASP.6 is still in beta, it stands to reason that there could be some performance fluctuations as it is iterating towards the final release.
How closely does your system platform match the one from the post on NVIDIA’s site?
3 times worst than the GPU results, I have the same V100 GPU. The CPU baseline is compatible with the data presented in the report.
Yikes! In that case one can only hope that someone from NVIDIA discovers this thread and follows up internally.