Head-to-head AI inference benchmark comparison of TPU7x (Google Ironwood) 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 103β390 tok/s/user interactivity band, at 174 tok/s/user on Kimi K3 2.8T: TPU7x runs 1960 tok/s/chip at $0.15/M tokens, Vera Rubin NVL72 runs 31257 at $0.03/M. Vera Rubin NVL72 is 355% cheaper per token; Vera Rubin NVL72 delivers 1494% more tok/s/chip.
Setting 246 tok/s/user as the target on Kimi K3 2.8T, TPU7x produces 1221 tok/s/chip ($0.23 per million tokens) and Vera Rubin NVL72 produces 21650 ($0.05). Vera Rubin NVL72 is 406% cheaper per token; Vera Rubin NVL72 delivers 1673% more tok/s/chip.
At 319 tok/s/user interactivity on Kimi K3 2.8T, TPU7x delivers 869 tok/s/chip at $0.33 per million tokens; Vera Rubin NVL72 delivers 13670 tok/s/chip at $0.07. Vera Rubin NVL72 is 349% cheaper per token; Vera Rubin NVL72 delivers 1474% more tok/s/chip at this point. (Numbers reflect the default agentic-traces Β· fp4 selection for this URL β table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/chip) | TPU7x:1960.3Vera Rubin NVL72:31256.9 | TPU7x:1220.9Vera Rubin NVL72:21650.1 | TPU7x:868.5Vera Rubin NVL72:13669.9 |
| Cost ($/M tok) | TPU7x:$0.146Vera Rubin NVL72:$0.032 | TPU7x:$0.234Vera Rubin NVL72:$0.046 | TPU7x:$0.329Vera Rubin NVL72:$0.073 |
| tok/s/MW | TPU7x:1624111Vera Rubin NVL72:9471776 | TPU7x:1011500Vera Rubin NVL72:6560628 | TPU7x:719591Vera Rubin NVL72:4142403 |
| Concurrency | TPU7x:~9Vera Rubin NVL72:~156 | TPU7x:~5Vera Rubin NVL72:~66 | TPU7x:~2Vera Rubin NVL72:~51 |
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.
