vLLM Optimizes Rubin: DeepSeek-V4.1-Flash Hits 58x GB300 Throughput

The vLLM team optimized DeepSeek-V4.1-Flash inference on NVIDIA Vera Rubin NVL72, achieving 1.7M tokens/s per GPU — 58x GB300 throughput on SemiAnalysis AgentX — aided by rack-level pooled KV cache with 96.5-97% hit rates.

2026-11-13 ~ 2026-11-13 · 4 related posts

• vLLM hits 58x GB300 throughput on Rubin with DeepSeek-V4.1-Flash optimization — vllm_project · 2026-11-13
• vLLM pools rack-scale KV cache on Rubin, holding 96.5-97% hit rate at high concurrency — vllm_project · 2026-11-13

2 near-duplicate retellings: vllm_project · vllm_project