DeepSeek R1 — B200 vs GB300 NVL72
Head-to-head AI inference benchmark comparison of B200 (NVIDIA Blackwell) and GB300 NVL72 (NVIDIA Blackwell) on DeepSeek R1. 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.
Setting 70 tok/s/user as the target on DeepSeek R1, B200 produces 1058 tok/s/GPU ($0.52 per million tokens) and GB300 NVL72 produces 5403 ($0.14). GB300 NVL72 is 279% cheaper per token; GB300 NVL72 delivers 411% more tok/s/GPU.
At 127 tok/s/user interactivity on DeepSeek R1, B200 delivers 275 tok/s/GPU at $1.97 per million tokens; GB300 NVL72 delivers 949 tok/s/GPU at $0.77. GB300 NVL72 is 157% cheaper per token; GB300 NVL72 delivers 245% more tok/s/GPU at this point.
B200 posts 158 tok/s/GPU for $3.40 per million tokens at 183 tok/s/user on DeepSeek R1; GB300 NVL72 posts 122 tok/s/GPU for $6.18. B200 is 82% cheaper per token; B200 delivers 30% more tok/s/GPU. (Numbers reflect the default 1k/1k · fp8 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/gpu) | B200:1058.4GB300 NVL72:5403.3 | B200:274.8GB300 NVL72:948.6 | B200:158.0GB300 NVL72:121.9 |
| Cost ($/M tok) | B200:$0.517GB300 NVL72:$0.136 | B200:$1.973GB300 NVL72:$0.767 | B200:$3.401GB300 NVL72:$6.176 |
| tok/s/MW | B200:487726GB300 NVL72:2573015 | B200:126650GB300 NVL72:451709 | B200:72795GB300 NVL72:58057 |
| Concurrency | B200:~62GB300 NVL72:~959 | B200:~8GB300 NVL72:~196 | B200:~23GB300 NVL72:~20 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.