vLLM optimizations on a 6-GPU cluster reduced voice AI latency by 40% for a Qwen-based system, enabling 500 concurrent sessions per node without hardware upgrades.

First public B200 PD disaggregation benchmark shows 7x token throughput improvement