Session 00 — From HPC to AI Infra: Parallel Computing and Parallel Programming
Why are supercomputing and AI infrastructure so closely related? Why have advances in compute enabled increasingly capable AI systems?
Large models run and train effectively on GPUs because, like scientific computing and climate simulation, their workloads can be highly parallel.
- Why can these workloads run in parallel?
- Where does parallelism appear?
- How do we express and execute parallel computation?
- Why does AI infrastructure require custom kernels, while ordinary CPU programming is rarely described in the same way?
This session develops the intuition needed to answer those questions.