DeepSeek R1 — B200 vs MI325X
Head-to-head AI inference benchmark comparison of B200 (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.
Near the low end of the 29–70 tok/s/user interactivity band, at 39 tok/s/user on DeepSeek R1: B200 runs 5078 tok/s/GPU at $0.11/M tokens, MI325X runs 341 at $1.03/M. B200 is 854% cheaper per token; B200 delivers 1387% more tok/s/GPU.
Setting 50 tok/s/user as the target on DeepSeek R1, B200 produces 3689 tok/s/GPU ($0.15 per million tokens) and MI325X produces 225 ($1.57). B200 is 966% cheaper per token; B200 delivers 1536% more tok/s/GPU.
At 60 tok/s/user interactivity on DeepSeek R1, B200 delivers 2573 tok/s/GPU at $0.21 per million tokens; MI325X delivers 132 tok/s/GPU at $2.67. B200 is 1162% cheaper per token; B200 delivers 1846% more tok/s/GPU at this point. (Numbers reflect the default 1k/1k · fp8 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/gpu) | B200:5078.3MI325X:341.4 | B200:3689.1MI325X:225.5 | B200:2572.7MI325X:132.2 |
| Cost ($/M tok) | B200:$0.108MI325X:$1.027 | B200:$0.147MI325X:$1.566 | B200:$0.212MI325X:$2.672 |
| tok/s/MW | B200:2340217MI325X:156612 | B200:1700029MI325X:103433 | B200:1185575MI325X:60648 |
| Concurrency | B200:~1746MI325X:~39 | B200:~1252MI325X:~19 | B200:~1011MI325X:~9 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.