MiniMax M3 428B — H200 vs MI325X Performance per Dollar
Cost per million tokens of H200 (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.
On MiniMax M3 428B at 51 tok/s/user, the per-million math comes out to $0.33 for H200 and $0.54 for MI325X; H200 delivers 61% more output per dollar.
At 95 tok/s/user on MiniMax M3 428B, H200 costs $0.59 per million tokens; MI325X costs $1.46. H200 is 145% more cost-efficient at this operating point.
H200 edges MI325X at 138 tok/s/user on MiniMax M3 428B — $0.92 per million tokens versus $2.63, a 186% cost-per-token gap. (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): H200 $1.22/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 | H200:$0.332MI325X:$0.535 | H200:$0.595MI325X:$1.459 | H200:$0.919MI325X:$2.626 |
| Concurrency | H200:~44MI325X:~49 | H200:~13MI325X:~9 | H200:~5MI325X:~4 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.