Head-to-head AI inference benchmark comparison of H100 (NVIDIA Hopper) and MI300X (AMD CDNA 3) on MiniMax M2.5/M2.7. 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.
H100 / MI300X on MiniMax M2.5/M2.7 at 37 tok/s/user: 2154 / 1530 tok/s/chip, $0.15 / $0.17 per million tokens. H100 is 14% cheaper per token; H100 delivers 41% more tok/s/chip.
Around the middle of the 21–84 tok/s/user interactivity band, at 53 tok/s/user on MiniMax M2.5/M2.7: H100 runs 1601 tok/s/chip at $0.20/M tokens, MI300X runs 1329 at $0.20/M. MI300X is 2% cheaper per token; H100 delivers 20% more tok/s/chip.
Setting 69 tok/s/user as the target on MiniMax M2.5/M2.7, H100 produces 1200 tok/s/chip ($0.27 per million tokens) and MI300X produces 1028 ($0.26). MI300X is 6% cheaper per token; H100 delivers 17% more tok/s/chip. (Numbers reflect the default 8k/1k · fp8 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/chip) | H100:2153.6MI300X:1529.5 | H100:1600.9MI300X:1328.8 | H100:1199.9MI300X:1028.5 |
| Cost ($/M tok) | H100:$0.151MI300X:$0.173 | H100:$0.203MI300X:$0.199 | H100:$0.271MI300X:$0.257 |
| tok/s/MW | H100:1571951MI300X:1100391 | H100:1168512MI300X:955941 | H100:875835MI300X:739905 |
| Concurrency | H100:~52MI300X:~12 | H100:~28MI300X:~5 | H100:~16MI300X:~4 |
Agentic inference metrics from the AgentX scenario and fixed-sequence inference metrics across models, hardware configurations, and serving parameters.
TCO $/chip/hr: VR200: 3.61H100: 1.17H200: 1.22B200: 1.73B300: 2.26GB200: 1.86GB300: 2.31MI300X: 0.95MI325X: 1.1MI355X: 1.5RTX6000PRO: 0.68Jalapeño (Teacup): 1.47
Source: SemiAnalysis Market July 2026 Pricing Surveys & AI Cloud TCO Model
Note: Disaggregated inference configurations (e.g., MoRI SGLang, Dynamo TRTLLM) calculate cost per million tokens per decode chip or per prefill chip, rather than per total chip count. This makes direct token cost comparison with aggregated configs not an apples-to-apples comparison.
Note: Disaggregated inference configurations (e.g., MoRI SGLang, Dynamo TRTLLM) calculate input throughput per decode chip or per prefill chip, rather than per total chip count. This makes direct input throughput comparison with aggregated configs not an apples-to-apples comparison.
Note: Disaggregated inference configurations (e.g., MoRI SGLang, Dynamo TRTLLM) calculate output throughput per decode chip or per prefill chip, rather than per total chip count. This makes direct output throughput comparison with aggregated configs not an apples-to-apples comparison.
Note: Disaggregated inference configurations (e.g., MoRI SGLang, Dynamo TRTLLM) calculate power per decode chip or per prefill chip, rather than per total chip count. This makes direct power comparison with aggregated configs not an apples-to-apples comparison.
Note: Disaggregated inference configurations (e.g., MoRI SGLang, Dynamo TRTLLM) calculate Joules per decode chip or per prefill chip, rather than per total chip count. This makes direct Joules per token comparison with aggregated configs not an apples-to-apples comparison.
No measurements to plot for this selection. Review the benchmark controls above or adjust quick filters.
Shift+Scroll to zoom • Drag to pan • Double-click to reset • Click a point to pin tooltip
MiniMax M2.5/2.7 230B 8K / 1K Total Tokens per $1 TCO vs. Interactivity