Scalable GNN and Computer‑Vision Training on Alvis for Large‑Scale Simulation & Learning
Dnr:
NAISS 2025/6-428
Type:
NAISS Medium Storage
Principal Investigator:
Jose Nunez-yanez
Affiliation:
Linköpings universitet
Start Date:
2026-01-01
End Date:
2026-12-01
Primary Classification:
10206: Computer Engineering
Webpage:
Allocation
Abstract
We request a NAISS Medium Compute allocation on Alvis to run GPU‑bound, simulation‑supported training of large‑scale graph neural networks and computer‑vision models using a containerized CUDA/NCCL stack (PyTorch, DGL/PyG). The Small cap is insufficient; our plan requires ~20,000 weighted GPU‑hours/month on Alvis for multi‑GPU experiments, ablations, and evaluation, with heavy data generation kept off Alvis.