DeepSeek R1 — GB300 NVL72 vs MI325X
Head-to-head AI inference benchmark comparison of GB300 NVL72 (NVIDIA Blackwell) and MI325X (AMD CDNA 3) 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 39 tok/s/user as the target on DeepSeek R1, GB300 NVL72 produces 7802 tok/s/GPU ($0.09 per million tokens) and MI325X produces 341 ($1.03). GB300 NVL72 is 997% cheaper per token; GB300 NVL72 delivers 2185% more tok/s/GPU.
At 50 tok/s/user interactivity on DeepSeek R1, GB300 NVL72 delivers 6920 tok/s/GPU at $0.11 per million tokens; MI325X delivers 225 tok/s/GPU at $1.57. GB300 NVL72 is 1364% cheaper per token; GB300 NVL72 delivers 2969% more tok/s/GPU at this point.
GB300 NVL72 posts 6170 tok/s/GPU for $0.12 per million tokens at 60 tok/s/user on DeepSeek R1; MI325X posts 132 tok/s/GPU for $2.67. GB300 NVL72 is 2133% cheaper per token; GB300 NVL72 delivers 4567% more tok/s/GPU. (Numbers reflect the default 1k/1k · fp8 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/gpu) | GB300 NVL72:7801.7MI325X:341.4 | GB300 NVL72:6919.7MI325X:225.5 | GB300 NVL72:6169.9MI325X:132.2 |
| Cost ($/M tok) | GB300 NVL72:$0.094MI325X:$1.027 | GB300 NVL72:$0.107MI325X:$1.566 | GB300 NVL72:$0.120MI325X:$2.672 |
| tok/s/MW | GB300 NVL72:3715112MI325X:156612 | GB300 NVL72:3295091MI325X:103433 | GB300 NVL72:2938041MI325X:60648 |
| Concurrency | GB300 NVL72:~2990MI325X:~39 | GB300 NVL72:~1960MI325X:~19 | GB300 NVL72:~1352MI325X:~9 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.