Head-to-head AI inference benchmark comparison of GB300 NVL72 (NVIDIA Blackwell) and H200 (NVIDIA Hopper) 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 43 tok/s/user on MiniMax M2.5/M2.7: GB300 NVL72 hits 10654 tok/s/chip, H200 hits 2816. Per-million costs land at $0.06 and $0.12 respectively. GB300 NVL72 is 100% cheaper per token; GB300 NVL72 delivers 278% more tok/s/chip.
GB300 NVL72 / H200 on MiniMax M2.5/M2.7 at 62 tok/s/user: 7309 / 2166 tok/s/chip, $0.09 / $0.16 per million tokens. GB300 NVL72 is 78% cheaper per token; GB300 NVL72 delivers 237% more tok/s/chip.
Toward the upper edge of the 24–100 tok/s/user interactivity band, at 81 tok/s/user on MiniMax M2.5/M2.7: GB300 NVL72 runs 3698 tok/s/chip at $0.17/M tokens, H200 runs 1595 at $0.21/M. GB300 NVL72 is 22% cheaper per token; GB300 NVL72 delivers 132% more tok/s/chip. (Numbers reflect the default 8k/1k · fp8 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/chip) | GB300 NVL72:10653.9H200:2816.4 | GB300 NVL72:7308.7H200:2165.7 | GB300 NVL72:3698.0H200:1594.8 |
| Cost ($/M tok) | GB300 NVL72:$0.060H200:$0.120 | GB300 NVL72:$0.088H200:$0.156 | GB300 NVL72:$0.174H200:$0.213 |
| tok/s/MW | GB300 NVL72:5025439H200:2055797 | GB300 NVL72:3447521H200:1580777 | GB300 NVL72:1744352H200:1164064 |
| Concurrency | GB300 NVL72:~200H200:~30 | GB300 NVL72:~197H200:~16 | GB300 NVL72:~51H200:~9 |
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
