AI Term of the Moment

high touch


Look Up Another Term


Definition: Vera Rubin


NVIDIA's flagship AI hardware as of 2026, superseding Blackwell. The name comes from American astronomer Vera Florence Cooper Rubin, who pioneered work on the rotation rates of galaxies. Vera is the CPU and Rubin is the GPU.

Often sold through system integrators, Vera CPUs can be purchased separately; however, Rubin GPUs are sold in various configurations. For example, the Vera Rubin Superchip includes one Vera CPU and two Rubin GPUs. HGX modules have eight GPUs and no CPUs, while the NVL72 rack includes 36 Vera CPUs, 72 Rubin GPUs and 20.7 terabytes of HBM4 memory and liquid cooling. The NVL144 doubles the number of chips. NVL is NVIDIA's high-speed interface between chips (see NVLink).




A Vera Rubin Rack
Selling for roughly $8 million, a small data center may have 10 to 100 NVL72 racks while a major AI factory may have thousands. At $8 million per rack, the total price tag can add up. (Image courtesy of NVIDIA Corporation.)






Different GPUs for Disaggregated Inference
To increase performance, "disaggregated inference" executes in two stages. To analyze the question, the "context" stage (prefill stage) requires a huge amount of computation (Rubin CPX GPUs). The "decode" stage (token generation) requires fast data transfer and high-speed memory (Rubin GPUs). See inference. (Image courtesy of NVIDIA.)





A
B
C
D
E
F
G
H
I
J
K
L
M
N
O
P
Q
R
S
T
U
V
W
X
Y
Z
misc