DeepSeek V4.1 Flash benchmark results on NVIDIA Vera Rubin NVL72

x.com1 min read

The first DeepSeek V4.1 Flash results on the NVIDIA Vera Rubin NVL72 showcase significant performance gains using Dynamo and vLLM. Compared to the GB300 NVL72 running the same stack, InferenceX performance metrics demonstrate a 63.4x total token throughput per chip at 400 tok/s/user, and a 93.3x increase at 600 tok/s/user. Open article →