NVIDIA Blackwell: B200 and GB200 NVL72, the rack-scale AI era
In one sentence At GTC 2024 NVIDIA announces Blackwell B200 (208B transistors, dual-die) and the GB200 NVL72 system (72 GPUs + 36 Grace CPUs in a rack). 30x faster inference for frontier LLMs.
Once a year NVIDIA holds its conference, GTC. In 2024 Jensen Huang takes the stage in his black leather jacket and announces Blackwell, the new AI chip family.
The headline number: the B200 chip has 208 billion transistors (Hopper H100 had 80). It is so big it is actually two chips stitched together. And the real product is not the single chip, it is the rack: the GB200 NVL72 is a 2-meter cabinet with 72 GPUs connected by internal NVLink, designed as "one giant computer".
For those training the biggest models (OpenAI, Anthropic, Meta, xAI, Google) it means: same model trained in half the time, or 4x larger models in the same time. The order queue forms instantly, with deliveries through 2025.
Companies
NVIDIA
Tools
B200, GB200 NVL72, NVLink
Tags
Sources