Nvidia sets new MLPerf records with 288 GPUs while AMD and Intel focus on different battles

MLCommons published MLPerf Inference v6.0 results on April 1, 2026; five new benchmarks were added.

Nvidia, AMD and Intel submitted results, but comparisons are only partially comparable due to differences in systems, models and scenarios.

Nvidia used 288‑GPU configurations in some tests (including DeepSeek‑R1 and GPT‑OSS‑120B) and was the only vendor to submit results for all new models and scenarios.

AMD compared against Nvidia B200 and B300 in single‑node eight‑GPU setups and did not submit DeepSeek‑R1 or Qwen3‑VL results; Intel focused on the workstation GPU market.

The GB300‑NVL72 with Blackwell Ultra GPUs achieved the highest throughput on the new workloads; Nvidia reported a 2.7× DeepSeek‑R1 throughput gain via software optimizations with Nebius and said software tweaks including Nvidia Dynamo cut token production costs by over 60%