MiniMax M3 428B — H100 vs MI325X Performance per Dollar
Cost per million tokens of H100 (NVIDIA Hopper) versus MI325X (AMD CDNA 3) on MiniMax M3 428B. 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.
At 57 tok/s/user on MiniMax M3 428B, H100 costs $0.42 per million tokens; MI325X costs $0.63. H100 is 50% more cost-efficient at this operating point.
H100 edges MI325X at 98 tok/s/user on MiniMax M3 428B — $0.91 per million tokens versus $1.53, a 68% cost-per-token gap.
Push MiniMax M3 428B to 140 tok/s/user and H100 lands at $1.39 per million tokens against MI325X's $2.69 — H100 pulls ahead by 94%. (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 · MI325X $1.10/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:$0.420MI325X:$0.631 | H100:$0.907MI325X:$1.527 | H100:$1.386MI325X:$2.688 |
| Concurrency | H100:~59MI325X:~38 | H100:~16MI325X:~9 | H100:~7MI325X:~3 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.