AMD and NVIDIA set new records in mlperf inference v6.1 benchmark

On September 16, 2026, Wccftech reported that MLPerf Inference v6.1 was released, with AMD and NVIDIA submitting results. MLPerf Inference is a benchmark suite maintained by MLCommons to evaluate AI inference performance across different hardware. AMD set a new record with 5.75 million tokens per second using 512 Instinct MI355X GPUs, marking its largest submission to date. NVIDIA submitted results for GB300 and GB200 with 288 GPUs, achieving 2.705 million tokens per second in offline mode and 2.028 million in server mode. NVIDIA's Vera Rubin VR200 also made its debut in the MLPerf benchmark, with 72 GPUs showing 95% faster performance than GB300. In the DeepSeek R1 test, AMD achieved 2.902 million tokens per second in offline mode and 2.405 million in server mode. NVIDIA's VR200, tested with 36 and 72 GPUs, outperformed the GB200 in both configurations. These results highlight the advancements in AI inference capabilities across major hardware vendors.