MiniMax M3 428B β H100 vs MI325X Performance per Dollar
Cost per million tokens of H100 (NVIDIA Hopper) versus MI325X (AMD CDNA 3) on MiniMax M3 428B. Large-hyperscaler-volume ownership TCO normalized by total tokens β 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 β same interactions as the main inference chart.
AgentX replays real coding-agent sessions rather than fixed-length prompts, so context grows turn over turn and most of each request is served from cache instead of being recomputed. That turns the comparison into a systems question: KV transfer between nodes, prefix-aware routing, and cache capacity all move the curve alongside raw chip throughput. Fixed-sequence workloads stay the clean baseline for kernel and silicon performance, so the two scenarios answer different questions about the same hardware. Learn more about AgentX β
At 13 tok/s/user on MiniMax M3 428B, H100 costs $0.50 per million tokens; MI325X costs $0.09. MI325X is 439% more cost-efficient at this operating point.
MI325X edges H100 at 15 tok/s/user on MiniMax M3 428B β $0.09 per million tokens versus $0.51, a 444% cost-per-token gap.
Push MiniMax M3 428B to 17 tok/s/user and H100 lands at $0.53 per million tokens against MI325X's $0.11 β MI325X pulls ahead by 376%. (Numbers reflect the default agentic-traces Β· fp8 selection for this URL β table and chart below update if you change sequence, precision, or model in the controls.)
Chip pricing (owning hyperscaler): H100 $1.17/chip/hr Β· 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:$0.497MI325X:$0.092 | H100:$0.510MI325X:$0.094 | H100:$0.532MI325X:$0.112 |
| Concurrency | H100:~2MI325X:~14 | H100:~1MI325X:~12 | H100:~1MI325X:~9 |
Inference Performance
Agentic inference metrics from the AgentX scenario and fixed-sequence inference metrics across models, hardware configurations, and serving parameters.