High-Performance Computing
High-performance computing marshals many processors together to solve problems too large or slow for a single machine.
When one computer is not enough
Full-fidelity fusion simulations — plasma turbulence, three-dimensional neutronics, coupled multiphysics — can require far more memory and arithmetic than any single processor offers. High-performance computing (HPC) divides such a problem across many processors working in concert.
How work is divided
- Domain decomposition: split the physical region into pieces, one per processor.
- Task parallelism: run independent calculations at the same time.
- Ensemble runs: sweep many parameter cases concurrently.
The communication cost
Processors working on adjacent regions must exchange boundary data, and this communication takes time. A well-designed parallel program keeps communication small relative to computation. When communication dominates, adding processors stops helping — and can even hurt.
Measuring scaling
Strong scaling asks how much faster a fixed problem runs on more processors; weak scaling asks whether a proportionally larger problem runs in the same time on proportionally more processors. Reporting both honestly is how HPC claims are made credible.
Why it matters here
HPC is what makes it feasible to explore a design space before hardware exists. Sweeping operating points for the breeder Hyperion or the D–³He burner, at fidelity high enough to trust, is a large computation that only becomes practical when spread across many processors.