DeepSeek R1 · Chip comparison
DeepSeek R1 — GB200 NVL72 vs MI325X
Head-to-head AI inference benchmark comparison of GB200 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.
Throughput at 49 tok/s/user on DeepSeek R1: GB200 NVL72 hits 6,529 tok/s/chip, MI325X hits 808. Costs per million tokens land at $0.079 and $0.38 respectively. GB200 NVL72 is 79% cheaper per token; GB200 NVL72 delivers 8.1× the tok/s/chip.
GB200 NVL72 / MI325X on DeepSeek R1 at 66 tok/s/user: 6,038 / 690 tok/s/chip, $0.086 / $0.44 per million tokens. GB200 NVL72 is 81% cheaper per token; GB200 NVL72 delivers 8.7× the tok/s/chip.
Toward the upper edge of the 32–100 tok/s/user interactivity band, at 83 tok/s/user on DeepSeek R1: GB200 NVL72 runs 5,432 tok/s/chip at $0.095 per million tokens, MI325X runs 570 at $0.54. GB200 NVL72 is 82% cheaper per token; GB200 NVL72 delivers 9.5× the tok/s/chip. (Numbers reflect the default 8k/1k · fp8 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
Interpolated from real benchmark data. Edit target interactivity values below to compare at different operating points.
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/chip) | GB200 NVL726,529MI325X808 | GB200 NVL726,038MI325X690 | GB200 NVL725,432MI325X570 |
| Cost ($/M tok) | GB200 NVL72$0.079MI325X$0.38 | GB200 NVL72$0.086MI325X$0.44 | GB200 NVL72$0.095MI325X$0.54 |
| tok/s/MW | GB200 NVL723,491,637MI325X478,325 | GB200 NVL723,228,815MI325X408,359 | GB200 NVL722,904,942MI325X337,068 |
| Concurrency | GB200 NVL72~1,994MI325X~15 | GB200 NVL72~939MI325X~9 | GB200 NVL72~437MI325X~6 |