Head-to-head AI inference benchmark comparison of GB300 NVL72 (NVIDIA Blackwell) and H200 (NVIDIA Hopper) on Kimi K3 2.8T. 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.
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 →
Near the low end of the 2–246 tok/s/user interactivity band, at 63 tok/s/user on Kimi K3 2.8T: GB300 NVL72 runs 11836 tok/s/chip at $0.05/M tokens, H200 runs 70 at $4.85/M. GB300 NVL72 is 8847% cheaper per token; GB300 NVL72 delivers 16840% more tok/s/chip.
Setting 124 tok/s/user as the target on Kimi K3 2.8T, GB300 NVL72 produces 7288 tok/s/chip ($0.09 per million tokens) and H200 produces 70 ($4.85). GB300 NVL72 is 5409% cheaper per token; GB300 NVL72 delivers 10330% more tok/s/chip.
At 185 tok/s/user interactivity on Kimi K3 2.8T, GB300 NVL72 delivers 3641 tok/s/chip at $0.18 per million tokens; H200 delivers 70 tok/s/chip at $4.85. GB300 NVL72 is 2652% cheaper per token; GB300 NVL72 delivers 5111% more tok/s/chip at this point. (Numbers reflect the default agentic-traces · fp4 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/chip) | GB300 NVL72:11836.4H200:69.9 | GB300 NVL72:7287.7H200:69.9 | GB300 NVL72:3641.1H200:69.9 |
| Cost ($/M tok) | GB300 NVL72:$0.054H200:$4.850 | GB300 NVL72:$0.088H200:$4.850 | GB300 NVL72:$0.176H200:$4.850 |
| tok/s/MW | GB300 NVL72:5583186H200:51001 | GB300 NVL72:3437578H200:51001 | GB300 NVL72:1717518H200:51001 |
| Concurrency | GB300 NVL72:~48H200:~1 | GB300 NVL72:~44H200:~1 | GB300 NVL72:~7H200:~1 |
Agentic inference metrics from the AgentX scenario and fixed-sequence inference metrics across models, hardware configurations, and serving parameters.
No measurements to plot for this selection. Review the benchmark controls above or adjust quick filters.
