vLLM Reveals DeepSeek V4.1 Flash Results on Vera Rubin NVL72

The vLLM team published first inference results for DeepSeek V4.1 Flash on NVIDIA's Vera Rubin NVL72 using Dynamo and the AgentX benchmark, showing up to 63x throughput gains and up to 93x per-chip throughput versus GB300.

2026-11-17 ~ 2026-11-17 · 3 related posts

1 near-duplicate retellings: vllm_project