DeepInfer beats vLLM on throughput with 8 B200 GPUs

Developers built DeepInfer-server and deployed DeepSeek-V4.1-Flash via GPUStack on a single server with eight Nvidia B200 GPUs in a DP8+EP8 setup. DeepInfer reported higher output throughput and lower inter-token latency, while vLLM had lower time to first token and better P95 latency.

• Tested on 8 B200 GPUs with DP8+EP8 and DeepSeek-V4.1-Flash
• DeepInfer led on output throughput and inter-token latency
• vLLM won on time to first token and P95 latency
• DeepInfer-server is a demo project, not DeepSeek's official service