Head-to-head AI inference benchmark comparison of GB300 NVL72 (NVIDIA Blackwell) and Vera Rubin NVL72 (NVIDIA Vera Rubin) on DeepSeekv4 Pro 0813 1.6T. 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 →
GB300 NVL72 posts 71122 tok/s/chip for $0.01 per million tokens at 130 tok/s/user on DeepSeekv4 Pro 0813 1.6T; Vera Rubin NVL72 posts 255517 tok/s/chip for $0.00. Vera Rubin NVL72 is 130% cheaper per token; Vera Rubin NVL72 delivers 259% more tok/s/chip.
Throughput at 187 tok/s/user on DeepSeekv4 Pro 0813 1.6T: GB300 NVL72 hits 13202 tok/s/chip, Vera Rubin NVL72 hits 136274. Per-million costs land at $0.05 and $0.01 respectively. Vera Rubin NVL72 is 561% cheaper per token; Vera Rubin NVL72 delivers 932% more tok/s/chip.
GB300 NVL72 / Vera Rubin NVL72 on DeepSeekv4 Pro 0813 1.6T at 243 tok/s/user: 5685 / 4625 tok/s/chip, $0.11 / $0.22 per million tokens. GB300 NVL72 is 92% cheaper per token; GB300 NVL72 delivers 23% more tok/s/chip. (Numbers reflect the default agentic-traces · fp4 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/chip) | GB300 NVL72:71121.8Vera Rubin NVL72:255516.7 | GB300 NVL72:13201.6Vera Rubin NVL72:136274.4 | GB300 NVL72:5685.3Vera Rubin NVL72:4625.1 |
| Cost ($/M tok) | GB300 NVL72:$0.009Vera Rubin NVL72:$0.004 | GB300 NVL72:$0.049Vera Rubin NVL72:$0.007 | GB300 NVL72:$0.113Vera Rubin NVL72:$0.217 |
| tok/s/MW | GB300 NVL72:33548002Vera Rubin NVL72:77429317 | GB300 NVL72:6227165Vera Rubin NVL72:41295259 | GB300 NVL72:2681731Vera Rubin NVL72:1401538 |
| Concurrency | GB300 NVL72:~325Vera Rubin NVL72:~2737 | GB300 NVL72:~32Vera Rubin NVL72:~1213 | GB300 NVL72:~1Vera Rubin NVL72:~19 |
Agentic inference metrics from the AgentX scenario and fixed-sequence inference metrics across models, hardware configurations, and serving parameters.
