TABLE II.
Estimated memory usage and resources allocated for the double branching microvessel shows that using the APR model shrinks the problem size from 235 nodes to one node on Summit.
| Model | Fluid Resolution | Num Fluid Pts | Memory | Resources |
|---|---|---|---|---|
| APR (window) | 0.5 μm | 8.1 × 107 | 33.0 GB | 3 NVIDIA V100 GPUs (fits on 1 node) |
| APR (bulk) | 2.5 μm | 4.4 × 108 | 183.0 GB | 2 IBM Power9 CPUs (fits on 1 node) |
| eFSI | 0.5 μm | 5.5 × 1010 | 22.5 TB | 1406 NVIDIA V100 GPUs (fits on 235 nodes) |