Head-to-head AI inference benchmark comparison of H200 (NVIDIA Hopper) and TPU7x (Google Ironwood) 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
Setting 99 tok/s/user as the target on Kimi K3 2.8T, H200 produces 70 tok/s/chip ($4.85 per million tokens) and TPU7x produces 4701 ($0.06). TPU7x is 7870% cheaper per token; TPU7x delivers 6629% more tok/s/chip.
At 196 tok/s/user interactivity on Kimi K3 2.8T, H200 delivers 70 tok/s/chip at $4.85 per million tokens; TPU7x delivers 1608 tok/s/chip at $0.18. TPU7x is 2626% cheaper per token; TPU7x delivers 2202% more tok/s/chip at this point.
H200 posts 70 tok/s/chip for $4.85 per million tokens at 293 tok/s/user on Kimi K3 2.8T; TPU7x posts 967 tok/s/chip for $0.30. TPU7x is 1539% cheaper per token; TPU7x delivers 1284% more tok/s/chip. (Numbers reflect the default agentic-traces Β· fp4 selection for this URL β table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/chip) | H200:69.9TPU7x:4701.4 | H200:69.9TPU7x:1608.1 | H200:69.9TPU7x:967.0 |
| Cost ($/M tok) | H200:$4.850TPU7x:$0.061 | H200:$4.850TPU7x:$0.178 | H200:$4.850TPU7x:$0.296 |
| tok/s/MW | H200:51001TPU7x:3895149 | H200:51001TPU7x:1332323 | H200:51001TPU7x:801140 |
| Concurrency | H200:~1TPU7x:~26 | H200:~1TPU7x:~7 | H200:~1TPU7x:~3 |
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.
