NVIDIA H200 Tensor Core

Core Architecture

Process Node: TSMC 4N (Custom 4nm)

Transistor Count: ~80 billion

Streaming Multiprocessors (SMs): Up to 144 (similar to H100, but with memory enhancements)

CUDA Cores: Not explicitly disclosed (Hopper focuses on Tensor Core efficiency)

Tensor Cores: 4th Gen with new FP8 acceleration

Thread Block Clustering: Allows better workload distribution across SMs