Background
MLPerf is a set of standard benchmarks for measuring how fast systems can train AI models. MLCommons publishes the results in rounds, which lets hardware makers and cloud providers compare their systems on the same tasks. Because the tests are fixed, the numbers describe time to reach a target quality, not peak chip specifications.
What happened
MLCommons released the MLPerf Training v6.1 results on November 16, 2026.[1] NVIDIA said it swept every benchmark in the round and was the only platform with results on every benchmark.[1] Google separately published its own results, with 10 MLPerf Training v6.1 results across seventh-generation Ironwood TPUs and NVIDIA Grace Blackwell GPUs.[2]
Key details
NVIDIA said its Vera Rubin NVL72 debut delivered up to 2x more performance per GPU than GB300 NVL72.[1] On the DeepSeek-V3 benchmark, NVIDIA said Vera Rubin NVL72 scaled from 256 to 1,024 GPUs at about 97% efficiency, cutting the time to train from 16 minutes to about four.[1]
In the same round, NVIDIA ran DeepSeek-V3 on 10,240 GB300 GPUs, and Microsoft Azure reported 10,240 GB200 GPUs at the same scale.[1] Scale here means the number of accelerators used in one run.
On the Google side, 8,192 Ironwood TPUs reached the pre-training target quality in 1.70 minutes. The run scaled to 4,096 chips with 96% efficiency.[2] In a separate run on agentic reinforcement learning, 640 Ironwood TPUs reached the target accuracy on a coding benchmark in 18.02 minutes.[2]
On Google Cloud A4X Max, 256 NVIDIA GB300 GPUs reached the pre-training target quality in 31.96 minutes.[2]
Impact
The results show two things at once: NVIDIA's new Vera Rubin system appearing in a public benchmark for the first time, and Google's TPU and GPU offerings competing on the same kind of training task. Because each company reports results on its own runs, readers should compare the numbers as each company's own measurements under the stated conditions.
Sources
- 1
- 2
Cite this page: MLPerf Training v6.1 results: NVIDIA Vera Rubin debut and Google Ironwood TPUs. TellWorld, https://www.tellworld.net/news/mlperf-training-v6-1-2026. Updated 2026-11-18