On-the-fly clustering for exascale molecular dynamics simulations. - presented by Dr Alizée Dubois and Thierry Carrard

On-the-fly clustering for exascale molecular dynamics simulations.

Alizée Dubois and Thierry Carrard

ADThierry Carrard
Slide at 24:28
Activités
Visionneur de documents
3 mars 15:25 0
37 sur68
pres_algo_killian.pdf
192,2%
mind
WEAK SCALING RESULTS
MD EAM
Analysis Ideal scaling
Per-node MPI/OMP work
distribution :
4 MPI X 32 OMP
WORKLOAD PER CORE :
160 . 103 atoms/core
113.103 voxels/core
1024
2048
4096
8192
16384
32768
Cores
81.9M atoms
5.2 B atoms
A. DUBOIS - T. CARRARD - COMPUTER PHYSICS COMMUNICATIONS SEMINAR SERIES - 03/03/25
II Votre écran est partagé par le biais de l'application app.zoom.us.
Arrêter le partage
Masquer
Share slide
Summary (AI generated)

For the weak scaling results, we maintain a constant load per core. As we increase the number of cores, we also increase the number of atoms, targeting approximately 160,000 atoms per core. Although the scalability of our existing system is not ideal, the scalability of the molecular dynamics is nearly perfect, indicating that communication is not a significant issue.

In our analysis, while the performance is not as strong, the bottleneck we identified—reducing the information from a single molecular interaction—is manageable. Our primary objective is to conduct an analysis that does not excessively increase computation time. We must ensure that the time required for analysis does not exceed that of a single step of molecular dynamics. The accompanying slide presents the ratio of analysis time to one step of dynamics.