MiniMax M2.5/M2.7 — H100 vs MI300X Performance per Dollar
Cost per million tokens of H100 (NVIDIA Hopper) versus MI300X (AMD CDNA 3) on MiniMax M2.5/M2.7. Owning-hyperscaler TCO normalized by output tokens — performance per dollar across LLM workloads. Pick the more cost-efficient SKU at every target interactivity level. Use the chart controls below to switch sequences, precisions, and metrics — same interactions as the main inference chart.
Push MiniMax M2.5/M2.7 to 54 tok/s/user and H100 lands at $0.45 per million tokens against MI300X's $0.27 — MI300X pulls ahead by 65%.
H100: $0.67 per million tokens. MI300X: $0.44. Both at 68 tok/s/user on MiniMax M2.5/M2.7, with MI300X 54% cheaper.
Toward the upper edge of the 41–95 tok/s/user interactivity band — at 82 tok/s/user — H100 runs $1.00 per million tokens on MiniMax M2.5/M2.7 while MI300X runs $0.75. MI300X is the cheaper choice by 34%. (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.)
Chip pricing (owning hyperscaler): H100 $1.17/chip/hr · MI300X $0.95/chip/hr. Source: SemiAnalysis Market July 2026 Pricing Surveys & AI Cloud TCO Model.
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Dollar per Million Tokens | H100:$0.448MI300X:$0.271 | H100:$0.673MI300X:$0.436 | H100:$1.003MI300X:$0.747 |
| Concurrency | H100:~53MI300X:~38 | H100:~29MI300X:~18 | H100:~16MI300X:~9 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.