we_are_coded.by CODE · The world, decoded
БГ

Who is who

Vera Rubin

NVIDIA's platform that follows GB300. It debuted in MLPerf on November 16, 2026, with leading numbers and one caveat: the Endpoints results are still pending verification.

Updated on 17 November 2026 · Checked on 17 November 2026

Vera Rubin is NVIDIA's next platform, the generation after GB300. Its debut in public benchmarks came on November 16, 2026, in the MLPerf Training v6.1 round and in NVIDIA's first MLPerf Endpoints submission.

The debut numbers, per NVIDIA: up to twice the performance per GPU against GB300 NVL72. On the DeepSeek-V3 test the system scales from 256 to 1,024 GPUs at about 97 percent efficiency, and that cuts training from 16 minutes to about 4.

Endpoints measures how much work a system does for every watt, because in a data center power is the ceiling. The first submission delivers more than 10 times the throughput per watt of the previous generation on agentic tasks. On the Kimi K3 test, a large open model, Vera Rubin NVL72 reaches the same response speed with 32 accelerators, while GB300 NVL72 needs 48.

Less hardware, more per watt. That is the shift NVIDIA is selling.

One note that keeps the numbers honest: Vera Rubin's Endpoints results are provisional. They are pending verification by MLCommons Association and may change after peer review. The GB300 results in the same round are already verified.

Primary sources

NVIDIA's blog

The visual is generated code-art, no third-party images.