DeepSeek R1 — B300 vs GB300 NVL72
Head-to-head AI inference benchmark comparison of B300 (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.
At 88 tok/s/user interactivity on DeepSeek R1, B300 delivers 4851 tok/s/GPU at $0.13 per million tokens; GB300 NVL72 delivers 10008 tok/s/GPU at $0.07. GB300 NVL72 is 85% cheaper per token; GB300 NVL72 delivers 106% more tok/s/GPU at this point.
B300 posts 542 tok/s/GPU for $1.18 per million tokens at 162 tok/s/user on DeepSeek R1; GB300 NVL72 posts 1967 tok/s/GPU for $0.38. GB300 NVL72 is 214% cheaper per token; GB300 NVL72 delivers 263% more tok/s/GPU.
Throughput at 235 tok/s/user on DeepSeek R1: B300 hits 234 tok/s/GPU, GB300 NVL72 hits 232. Per-million costs land at $2.77 and $3.27 respectively. B300 is 18% cheaper per token; throughput per GPU is essentially tied. (Numbers reflect the default 1k/1k · fp4 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/gpu) | B300:4850.6GB300 NVL72:10008.0 | B300:541.8GB300 NVL72:1966.9 | B300:234.3GB300 NVL72:232.4 |
| Cost ($/M tok) | B300:$0.134GB300 NVL72:$0.073 | B300:$1.182GB300 NVL72:$0.377 | B300:$2.765GB300 NVL72:$3.275 |
| tok/s/MW | B300:2235282GB300 NVL72:4765719 | B300:249692GB300 NVL72:936607 | B300:107973GB300 NVL72:110644 |
| Concurrency | B300:~386GB300 NVL72:~1716 | B300:~73GB300 NVL72:~286 | B300:~22GB300 NVL72:~18 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.