Scientific Computing
High PerformanceComputing (HPC)
Under DevelopmentParallelize computations across CUDA-enabled GPU clusters, manage job queues using Slurm, and optimize message passing parameters (MPI).
Planetary Compute Infrastructure
Platform VELSTROM provides access to dedicated high-performance clusters. This hardware allows developers and researchers to execute complex physics calculations, mesh millions of geometry elements, and compute time-series remote sensing arrays in parallel.
Slurm Job Queue Submission
Computations are scheduled using Slurm queues. Submit tasks via shell scripts outlining CPU, memory, and GPU allocations.
submit_job.sh
| 1 | #!/bin/bash |
| 2 | #SBATCH --job-name=velstrom_fem_sim |
| 3 | #SBATCH --output=logs/sim_%j.log |
| 4 | #SBATCH --nodes=4 # Allocate 4 compute nodes |
| 5 | #SBATCH --ntasks-per-node=16 # 16 parallel tasks per node (64 total) |
| 6 | #SBATCH --gres=gpu:tesla_h100:2 # Request 2 NVIDIA H100 GPUs per node |
| 7 | #SBATCH --time=04:00:00 # 4-hour runtime limit |
| 8 | #SBATCH --partition=planetary_compute |
| 9 | |
| 10 | # Run parallel finite-element fluid simulation using MPI and OpenFOAM |
| 11 | mpirun -np 64 simpleFoam -parallel |
Simulation Benchmarks
| Model Identifier | Hardware Target | Grid Mesh Elements | Solve Time | Parallel Efficiency |
|---|---|---|---|---|
| Hydrology_Runoff_2D | 16x AMD EPYC (256 Cores) | 12.4 Million | 14m 22s | 94.2% |
| Micromagnetics_ThinFilm | 2x NVIDIA H100 (Cuda 12) | 6.8 Million | 4m 12s | 98.1% |
| Soil_Carbon_Flow_3D | 4x AMD EPYC (64 Cores) | 2.1 Million | 8m 45s | 89.4% |