The previous GPU generation from NVIDIA. Although superseded by Vera Rubin, Blackwell chips are active products estimated to be available through 2027.
In 2024, NVIDIA introduced the Blackwell B100 and B200, each GPU having two dies linked together with 192GB of HBM3e memory and 208 billion transistors on a single chip. Named after the distinguished American mathematician David Blackwell, the B200 is roughly 25% faster than the B100 and about 30 times faster the NVIDIA H100. See
H100,
A100,
GPU and
high bandwidth memory.
GB200 - Grace Blackwell Superchip
Superseding the Grace Hopper Superchip, the Grace Blackwell Superchip links two B200 GPUs and one Grace CPU via NVLink at 900 Gbps. The B200 GPU is 30 times faster than NVIDIA's H100 Tensor Core GPU and reduces energy up to 25 times. See
Grace Hopper Superchip and
DGX Spark.
GB10 - A Mini Version
In 2025, NVIDIA launched a lightweight GB200 for use in its desktop AI computer (see
DGX Spark).
Not Your Gamer's GPU
The Blackwell B200 costs around $30,000. A B200 server with 16 GPUs is approximately $240,000, and a GB200 NVL72 server costs around $3 million. No wonder an AI data center can cost $3 billion to build and populate with equipment. See
AI data center and
rack mounted.
Blackwell Ultra and Vera Rubin
Launched in 2025, Blackwell Ultra is an enhanced version of Blackwell with more memory. The successor to Blackwell is Rubin, an AI platform that includes the Vera CPU and Rubin GPU. See
Vera Rubin.