NVIDIA Hopper Architecture In-Depth

Originally published at: NVIDIA Hopper Architecture In-Depth | NVIDIA Technical Blog

Everything you want to know about the new H100 GPU.

1 Like

A question, arose with reading the news, is related to L2 (L1) bandwidth for sustained 3TB/s memory bandwidth on that latency levels of Grace and Hopper architectures.
Are there suggestions or real numbers (from cpu/gpu analysis) for to sort in these high numbers on cache and memory transfer bw (thx)?