MLPerf Training 6.1: Nvidia Rubin and Google Ironwood debut, AI post-training gets a benchmark — what the results show
MLCommons' November 16 MLPerf Training round added the first post-training (GRPO reinforcement learning) workload and the first results from Nvidia's Vera Rubin and Google's Ironwood TPU. Here's what the numbers say, which claims are the vendors' own, and how to read them.