Head-to-head AI inference benchmark comparison of GB300 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
Setting 183 tok/s/user as the target on Kimi K3 2.8T, GB300 NVL72 produces 8346 tok/s/chip ($0.08 per million tokens) and Vera Rubin NVL72 produces 29856 ($0.03). Vera Rubin NVL72 is 129% cheaper per token; Vera Rubin NVL72 delivers 258% more tok/s/chip.
At 263 tok/s/user interactivity on Kimi K3 2.8T, GB300 NVL72 delivers 4174 tok/s/chip at $0.15 per million tokens; Vera Rubin NVL72 delivers 19637 tok/s/chip at $0.05. Vera Rubin NVL72 is 201% cheaper per token; Vera Rubin NVL72 delivers 370% more tok/s/chip at this point.
GB300 NVL72 posts 2273 tok/s/chip for $0.28 per million tokens at 343 tok/s/user on Kimi K3 2.8T; Vera Rubin NVL72 posts 11732 tok/s/chip for $0.09. Vera Rubin NVL72 is 230% cheaper per token; Vera Rubin NVL72 delivers 416% more tok/s/chip. (Numbers reflect the default agentic-traces Β· fp4 selection for this URL β table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/chip) | GB300 NVL72:8346.2Vera Rubin NVL72:29856.3 | GB300 NVL72:4174.3Vera Rubin NVL72:19636.8 | GB300 NVL72:2272.8Vera Rubin NVL72:11731.8 |
| Cost ($/M tok) | GB300 NVL72:$0.077Vera Rubin NVL72:$0.034 | GB300 NVL72:$0.154Vera Rubin NVL72:$0.051 | GB300 NVL72:$0.282Vera Rubin NVL72:$0.085 |
| tok/s/MW | GB300 NVL72:3936870Vera Rubin NVL72:9047368 | GB300 NVL72:1969030Vera Rubin NVL72:5950538 | GB300 NVL72:1072072Vera Rubin NVL72:3555091 |
| Concurrency | GB300 NVL72:~27Vera Rubin NVL72:~144 | GB300 NVL72:~18Vera Rubin NVL72:~60 | GB300 NVL72:~9Vera Rubin NVL72:~48 |
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.
