DeepSeek R1 — B300 vs H100
Head-to-head AI inference benchmark comparison of B300 (NVIDIA Blackwell) and H100 (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.
Near the low end of the 9–128 tok/s/user interactivity band, at 38 tok/s/user on DeepSeek R1: B300 runs 4077 tok/s/GPU at $0.16/M tokens, H100 runs 276 at $1.32/M. B300 is 728% cheaper per token; B300 delivers 1376% more tok/s/GPU.
Setting 68 tok/s/user as the target on DeepSeek R1, B300 produces 1277 tok/s/GPU ($0.54 per million tokens) and H100 produces 87 ($4.18). B300 is 672% cheaper per token; B300 delivers 1371% more tok/s/GPU.
At 99 tok/s/user interactivity on DeepSeek R1, B300 delivers 476 tok/s/GPU at $1.37 per million tokens; H100 delivers 23 tok/s/GPU at $15.6. B300 is 1044% cheaper per token; B300 delivers 1940% more tok/s/GPU at this point. (Numbers reflect the default 1k/1k · fp8 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/gpu) | B300:4077.1H100:276.2 | B300:1277.2H100:86.8 | B300:476.3H100:23.3 |
| Cost ($/M tok) | B300:$0.159H100:$1.320 | B300:$0.542H100:$4.182 | B300:$1.367H100:$15.632 |
| tok/s/MW | B300:1878829H100:159663 | B300:588571H100:50179 | B300:219493H100:13494 |
| Concurrency | B300:~2684H100:~640 | B300:~457H100:~68 | B300:~103H100:~8 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.