AgentX · DeepSeek-V4.1-Flash

Vera Rubin NVL72 vs GB300 NVL72 throughput per GPU

AgentX workload, vLLM. Total tokens/s per GPU against p90 generation tokens/s per user (up and right is better).