Head-to-head AI inference benchmark comparison of GB200 NVL72 (NVIDIA Blackwell) and H200 (NVIDIA Hopper) on gpt-oss 120B. 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.
Setting 115 tok/s/user as the target on gpt-oss 120B, GB200 NVL72 produces 42256 tok/s/chip ($0.01 per million tokens) and H200 produces 5691 ($0.06). GB200 NVL72 is 387% cheaper per token; GB200 NVL72 delivers 642% more tok/s/chip.
At 162 tok/s/user interactivity on gpt-oss 120B, GB200 NVL72 delivers 32452 tok/s/chip at $0.02 per million tokens; H200 delivers 4014 tok/s/chip at $0.08. GB200 NVL72 is 430% cheaper per token; GB200 NVL72 delivers 708% more tok/s/chip at this point.
GB200 NVL72 posts 25143 tok/s/chip for $0.02 per million tokens at 209 tok/s/user on gpt-oss 120B; H200 posts 2541 tok/s/chip for $0.13. GB200 NVL72 is 549% cheaper per token; GB200 NVL72 delivers 889% more tok/s/chip. (Numbers reflect the default 8k/1k · fp4 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/chip) | GB200 NVL72:42255.8H200:5691.4 | GB200 NVL72:32452.2H200:4014.4 | GB200 NVL72:25143.3H200:2541.4 |
| Cost ($/M tok) | GB200 NVL72:$0.012H200:$0.060 | GB200 NVL72:$0.016H200:$0.084 | GB200 NVL72:$0.021H200:$0.133 |
| tok/s/MW | GB200 NVL72:22596704H200:4154336 | GB200 NVL72:17354104H200:2930201 | GB200 NVL72:13445593H200:1855043 |
| Concurrency | GB200 NVL72:~393H200:~16 | GB200 NVL72:~253H200:~13 | GB200 NVL72:~84H200:~5 |
Agentic inference metrics from the AgentX scenario and fixed-sequence inference metrics across models, hardware configurations, and serving parameters.
TCO $/chip/hr:
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
gpt-oss 120B 8K / 1K Total Tokens per $1 TCO vs. Interactivity