DeepSeek V4 Pro 1.6T — B200 vs GB300 NVL72
Head-to-head AI inference benchmark comparison of B200 (NVIDIA Blackwell) and GB300 NVL72 (NVIDIA Blackwell) on DeepSeek V4 Pro 1.6T. Latency, throughput, and cost across LLM workloads. Use the chart controls below to switch sequences, precisions, and metrics — same interactions as the main inference chart.
Throughput at 50 tok/s/user on DeepSeek V4 Pro 1.6T: B200 hits 2373 tok/s/GPU, GB300 NVL72 hits 7507. Per-million costs land at $0.23 and $0.10 respectively. GB300 NVL72 is 131% cheaper per token; GB300 NVL72 delivers 216% more tok/s/GPU.
B200 / GB300 NVL72 on DeepSeek V4 Pro 1.6T at 88 tok/s/user: 740 / 2508 tok/s/GPU, $0.73 / $0.29 per million tokens. GB300 NVL72 is 149% cheaper per token; GB300 NVL72 delivers 239% more tok/s/GPU.
Toward the upper edge of the 13–163 tok/s/user interactivity band, at 126 tok/s/user on DeepSeek V4 Pro 1.6T: B200 runs 458 tok/s/GPU at $1.19/M tokens, GB300 NVL72 runs 923 at $0.78/M. GB300 NVL72 is 52% cheaper per token; GB300 NVL72 delivers 101% more tok/s/GPU. (Numbers reflect the default 8k/1k · fp4 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/gpu) | B200:2373.1GB300 NVL72:7506.7 | B200:740.4GB300 NVL72:2508.0 | B200:458.2GB300 NVL72:923.0 |
| Cost ($/M tok) | B200:$0.227GB300 NVL72:$0.098 | B200:$0.732GB300 NVL72:$0.294 | B200:$1.191GB300 NVL72:$0.785 |
| tok/s/MW | B200:1093608GB300 NVL72:3574596 | B200:341207GB300 NVL72:1194304 | B200:211149GB300 NVL72:439516 |
| Concurrency | B200:~51GB300 NVL72:~989 | B200:~8GB300 NVL72:~254 | B200:~4GB300 NVL72:~43 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.