DeepSeek V4 Pro 1.6T — B200 vs GB200 NVL72
Head-to-head AI inference benchmark comparison of B200 (NVIDIA Blackwell) and GB200 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.
B200 posts 4023 tok/s/GPU for $0.14 per million tokens at 55 tok/s/user on DeepSeek V4 Pro 1.6T; GB200 NVL72 posts 9497 tok/s/GPU for $0.06. GB200 NVL72 is 110% cheaper per token; GB200 NVL72 delivers 136% more tok/s/GPU.
Throughput at 105 tok/s/user on DeepSeek V4 Pro 1.6T: B200 hits 976 tok/s/GPU, GB200 NVL72 hits 6644. Per-million costs land at $0.56 and $0.09 respectively. GB200 NVL72 is 503% cheaper per token; GB200 NVL72 delivers 581% more tok/s/GPU.
B200 / GB200 NVL72 on DeepSeek V4 Pro 1.6T at 154 tok/s/user: 618 / 526 tok/s/GPU, $0.88 / $1.17 per million tokens. B200 is 33% cheaper per token; B200 delivers 17% more tok/s/GPU. (Numbers reflect the default 8k/1k · fp4 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/gpu) | B200:4022.6GB200 NVL72:9497.1 | B200:975.6GB200 NVL72:6644.2 | B200:617.6GB200 NVL72:525.8 |
| Cost ($/M tok) | B200:$0.136GB200 NVL72:$0.065 | B200:$0.559GB200 NVL72:$0.093 | B200:$0.879GB200 NVL72:$1.169 |
| tok/s/MW | B200:1853711GB200 NVL72:4522446 | B200:449587GB200 NVL72:3163896 | B200:284603GB200 NVL72:250366 |
| Concurrency | B200:~91GB200 NVL72:~16349 | B200:~9GB200 NVL72:~2152 | B200:~4GB200 NVL72:~45 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.