Computing · January 2026

Nvidia Vera Rubin

Six new chips designed as one rack-scale AI computer.

Nvidia launched the Vera Rubin platform at CES in full production: a Rubin GPU rated at 50 petaflops of NVFP4 and an NVL72 rack with 260 TB/s of NVLink bandwidth. Nvidia claimed up to 10 times lower cost per inference token than Blackwell.

See it in the living tree Wikipedia Source

Arcs above join it to what it stood on, arcs below to what it led to. The bar places this stretch on the tree's whole timeline.

What it built on 2

Nvidia Blackwell 2024
High Bandwidth Memory 2013

Related measurements

The dashed line marks when it first worked.

ML hardware FLOP/s per dollar Source: Epoch AI, ML hardware (CC BY 4.0)

Training compute of notable AI models Source: Epoch AI, Notable AI Models (CC BY 4.0)

Transistors per microprocessor Latest: 58.2B transistors (2021) Source: Our World in Data