Head-to-head AI inference benchmark comparison of GB200 NVL72 (NVIDIA Blackwell) and MI355X (AMD CDNA 4) 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.
Throughput at 50 tok/s/user on MiniMax M2.5/M2.7: GB200 NVL72 hits 8615 tok/s/chip, MI355X hits 4020. Per-million costs land at $0.06 and $0.10 respectively. GB200 NVL72 is 73% cheaper per token; GB200 NVL72 delivers 114% more tok/s/chip.
GB200 NVL72 / MI355X on MiniMax M2.5/M2.7 at 70 tok/s/user: 4701 / 2713 tok/s/chip, $0.11 / $0.15 per million tokens. GB200 NVL72 is 40% cheaper per token; GB200 NVL72 delivers 73% more tok/s/chip.
Toward the upper edge of the 31–110 tok/s/user interactivity band, at 91 tok/s/user on MiniMax M2.5/M2.7: GB200 NVL72 runs 2725 tok/s/chip at $0.19/M tokens, MI355X runs 1744 at $0.24/M. GB200 NVL72 is 26% cheaper per token; GB200 NVL72 delivers 56% more tok/s/chip. (Numbers reflect the default 8k/1k · fp8 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/chip) | GB200 NVL72:8615.0MI355X:4020.4 | GB200 NVL72:4701.4MI355X:2712.7 | GB200 NVL72:2725.1MI355X:1743.8 |
| Cost ($/M tok) | GB200 NVL72:$0.060MI355X:$0.104 | GB200 NVL72:$0.110MI355X:$0.154 | GB200 NVL72:$0.190MI355X:$0.239 |
| tok/s/MW | GB200 NVL72:4606957MI355X:1923656 | GB200 NVL72:2514125MI355X:1297935 | GB200 NVL72:1457292MI355X:834337 |
| Concurrency | GB200 NVL72:~130MI355X:~22 | GB200 NVL72:~49MI355X:~9 | GB200 NVL72:~22MI355X:~5 |
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.
Matching measurements exist, but their chip series are hidden.
Shift+Scroll to zoom • Drag to pan • Double-click to reset • Click a point to pin tooltip
1 The ATOM engine is promising, however it has yet to serve production tokens. It is still in its infant stage.
