DeepSeek R1 — B300 vs H200
Head-to-head AI inference benchmark comparison of B300 (NVIDIA Blackwell) and H200 (NVIDIA Hopper) 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 50 tok/s/user as the target on DeepSeek R1, B300 produces 5852 tok/s/GPU ($0.11 per million tokens) and H200 produces 1615 ($0.24). B300 is 115% cheaper per token; B300 delivers 262% more tok/s/GPU.
At 89 tok/s/user interactivity on DeepSeek R1, B300 delivers 2497 tok/s/GPU at $0.26 per million tokens; H200 delivers 569 tok/s/GPU at $0.69. B300 is 165% cheaper per token; B300 delivers 339% more tok/s/GPU at this point.
B300 posts 878 tok/s/GPU for $0.75 per million tokens at 128 tok/s/user on DeepSeek R1; H200 posts 292 tok/s/GPU for $1.33. B300 is 77% cheaper per token; B300 delivers 201% more tok/s/GPU. (Numbers reflect the default 8k/1k · fp8 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/gpu) | B300:5852.2H200:1615.3 | B300:2496.6H200:569.0 | B300:878.4H200:292.1 |
| Cost ($/M tok) | B300:$0.112H200:$0.241 | B300:$0.260H200:$0.689 | B300:$0.750H200:$1.329 |
| tok/s/MW | B300:2696874H200:933717 | B300:1150501H200:328875 | B300:404792H200:168837 |
| Concurrency | B300:~288H200:~74 | B300:~58H200:~6 | B300:~9H200:~21 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.