NVIDIA Hopper GPU

Realizes: Transformer and dense matrix acceleration

The Hopper family (H100) is NVIDIA's GPU architecture for large-scale transformer training, pairing a new Transformer Engine with CUDA/SIMT cores and tensor cores on a 4nm/5nm FinFET node; HGX H100 cabinets tie multiple Hopper GPUs via NVLink to deliver deterministic throughput for massive AI workloads.

Examples

NVIDIA DGX H100

The DGX H100 appliance binds eight H100 SXM modules via the HGX H100 fabric, delivering deterministic transformer training and inference through the Transformer Engine and NVLink-accelerated SIMT pipelines.

[MIXED_PRECISION_MATMUL,TRANSFORMER_ENGINE] nanoseconds large pJ