Landmark AI Infrastructure · 1 min read
NVIDIA H100 and Hopper architecture: the foundation-model GPU
In one sentence At GTC 2022 NVIDIA unveils the Hopper architecture and the H100 GPU, with FP8 Transformer Engine and NVLink 4. It will become the hardware substrate for nearly every large LLM of the following years.
Reading level
NVIDIA shows a new generation of graphics processors designed not for video games but to train AIs. They're called H100.
The big novelty is a piece of the chip — the "Transformer Engine" — built specifically to run GPT-style models faster and with less memory.
From now on, when you hear "we trained the model on X thousand H100 GPUs", those are these. They become AI's hard currency: H100 clusters are what separates who can build a foundation model from who can't.
Companies
NVIDIA
Tools
H100
Tags
NVIDIAH100HopperGPUTransformer Engine
Sources