DeepSeek R1 · Chip comparison
DeepSeek R1 — GB200 NVL72 vs MI300X
Head-to-head AI inference benchmark comparison of GB200 NVL72 (NVIDIA Blackwell) and MI300X (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.
GB200 NVL72 / MI300X on DeepSeek R1 at 37 tok/s/user: 7,073 / 531 tok/s/chip, $0.073 / $0.50 per million tokens. GB200 NVL72 is 85% cheaper per token; GB200 NVL72 delivers 13× the tok/s/chip.
Around the middle of the 32–54 tok/s/user interactivity band, at 43 tok/s/user on DeepSeek R1: GB200 NVL72 runs 6,988 tok/s/chip at $0.074 per million tokens, MI300X runs 425 at $0.62. GB200 NVL72 is 88% cheaper per token; GB200 NVL72 delivers 16× the tok/s/chip.
Setting 49 tok/s/user as the target on DeepSeek R1, GB200 NVL72 produces 6,529 tok/s/chip ($0.079 per million tokens) and MI300X produces 321 ($0.82). GB200 NVL72 is 90% cheaper per token; GB200 NVL72 delivers 20× the tok/s/chip. (Numbers reflect the default 8k/1k · fp8 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
Interpolated from real benchmark data. Edit target interactivity values below to compare at different operating points.
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/chip) | GB200 NVL727,073MI300X531 | GB200 NVL726,988MI300X425 | GB200 NVL726,529MI300X321 |
| Cost ($/M tok) | GB200 NVL72$0.073MI300X$0.50 | GB200 NVL72$0.074MI300X$0.62 | GB200 NVL72$0.079MI300X$0.82 |
| tok/s/MW | GB200 NVL723,782,124MI300X381,904 | GB200 NVL723,736,945MI300X305,827 | GB200 NVL723,491,637MI300X230,743 |
| Concurrency | GB200 NVL72~2,484MI300X~13 | GB200 NVL72~4,883MI300X~9 | GB200 NVL72~1,994MI300X~6 |