Hardware guide

NVIDIA and Apple hardware for local and production inference.

H100 SXM: The course's production reference accelerator: fast HBM, a server-only form factor, and the baseline for memory-bound decode math.

HGX H200: An eight-GPU Hopper baseboard for models and serving workloads that need more memory and bandwidth than one accelerator provides.

DGX B200: A complete eight-GPU Blackwell system for organizations operating large training and inference workloads as shared infrastructure.