Head-to-head AI inference benchmark comparison of GB200 NVL72 (NVIDIA Blackwell) and Vera Rubin NVL72 (NVIDIA Vera Rubin) on Kimi K3 2.8T. Latency, throughput, and cost across LLM workloads. Use the chart controls below to switch sequences, precisions, and metrics β 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 β
Near the low end of the 103β218 tok/s/user interactivity band, at 131 tok/s/user on Kimi K3 2.8T: GB200 NVL72 runs 3127 tok/s/chip at $0.17/M tokens, Vera Rubin NVL72 runs 39019 at $0.03/M. Vera Rubin NVL72 is 543% cheaper per token; Vera Rubin NVL72 delivers 1148% more tok/s/chip.
Setting 161 tok/s/user as the target on Kimi K3 2.8T, GB200 NVL72 produces 2529 tok/s/chip ($0.20 per million tokens) and Vera Rubin NVL72 produces 33434 ($0.03). Vera Rubin NVL72 is 581% cheaper per token; Vera Rubin NVL72 delivers 1222% more tok/s/chip.
At 190 tok/s/user interactivity on Kimi K3 2.8T, GB200 NVL72 delivers 1414 tok/s/chip at $0.37 per million tokens; Vera Rubin NVL72 delivers 28822 tok/s/chip at $0.03. Vera Rubin NVL72 is 950% cheaper per token; Vera Rubin NVL72 delivers 1938% more tok/s/chip at this point. (Numbers reflect the default agentic-traces Β· fp4 selection for this URL β table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/chip) | GB200 NVL72:3126.9Vera Rubin NVL72:39019.4 | GB200 NVL72:2529.1Vera Rubin NVL72:33433.8 | GB200 NVL72:1414.0Vera Rubin NVL72:28821.7 |
| Cost ($/M tok) | GB200 NVL72:$0.165Vera Rubin NVL72:$0.026 | GB200 NVL72:$0.204Vera Rubin NVL72:$0.030 | GB200 NVL72:$0.365Vera Rubin NVL72:$0.035 |
| tok/s/MW | GB200 NVL72:1672152Vera Rubin NVL72:11824071 | GB200 NVL72:1352480Vera Rubin NVL72:10131449 | GB200 NVL72:756162Vera Rubin NVL72:8733834 |
| Concurrency | GB200 NVL72:~12Vera Rubin NVL72:~215 | GB200 NVL72:~10Vera Rubin NVL72:~174 | GB200 NVL72:~2Vera Rubin NVL72:~133 |
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.
