Core concept
Main + worker nodes. Main runs coordination. Workers execute compute. All connected via placement group + EFA for low-latency.
Advertisement
How it works
EFA (Elastic Fabric Adapter) provides low-latency, high-bandwidth networking. Essential for multi-node MPI.
Advertisement
Trade-offs + gotchas
Molecular dynamics. Weather simulation. Multi-node ML training (small scale). Any tightly-coupled HPC.