DeepSeek R1 · GPU comparison

DeepSeek R1 — GB200 NVL72 vs GB300 NVL72

Head-to-head AI inference benchmark comparison of GB200 NVL72 (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 85 tok/s/user as the target on DeepSeek R1, GB200 NVL72 produces 7753 tok/s/GPU ($0.08 per million tokens) and GB300 NVL72 produces 10422 ($0.07). GB300 NVL72 is 13% cheaper per token; GB300 NVL72 delivers 34% more tok/s/GPU.

At 152 tok/s/user interactivity on DeepSeek R1, GB200 NVL72 delivers 1361 tok/s/GPU at $0.45 per million tokens; GB300 NVL72 delivers 2644 tok/s/GPU at $0.28. GB300 NVL72 is 61% cheaper per token; GB300 NVL72 delivers 94% more tok/s/GPU at this point.

GB200 NVL72 posts 291 tok/s/GPU for $2.11 per million tokens at 219 tok/s/user on DeepSeek R1; GB300 NVL72 posts 282 tok/s/GPU for $2.62. GB200 NVL72 is 24% cheaper per token; GB200 NVL72 delivers 3% more tok/s/GPU. (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.)

View performance-per-dollar view →

Interpolated from real benchmark data. Edit target interactivity values below to compare at different operating points.
Metric
Interactivity (tok/s/user)
Interactivity (tok/s/user)
Interactivity (tok/s/user)
Throughput (tok/s/gpu)
GB200 NVL72:7753.2GB300 NVL72:10421.6
GB200 NVL72:1360.9GB300 NVL72:2644.5
GB200 NVL72:290.7GB300 NVL72:281.6
Cost ($/M tok)
GB200 NVL72:$0.079GB300 NVL72:$0.070
GB200 NVL72:$0.447GB300 NVL72:$0.277
GB200 NVL72:$2.110GB300 NVL72:$2.617
tok/s/MW
GB200 NVL72:3691993GB300 NVL72:4962665
GB200 NVL72:648067GB300 NVL72:1259271
GB200 NVL72:138416GB300 NVL72:134096
Concurrency
GB200 NVL72:~2246GB300 NVL72:~1780
GB200 NVL72:~226GB300 NVL72:~406
GB200 NVL72:~27GB300 NVL72:~24

Inference Performance

Inference performance metrics across different models, hardware configurations, and serving parameters.