Head-to-head AI inference benchmark comparison of GB300 NVL72 (NVIDIA Blackwell) and H100 (NVIDIA Hopper) on Muse Spark Open 816B. 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
GB300 NVL72 posts 22642 tok/s/chip for $0.03 per million tokens at 51 tok/s/user on Muse Spark Open 816B; H100 posts 1316 tok/s/chip for $0.25. GB300 NVL72 is 772% cheaper per token; GB300 NVL72 delivers 1621% more tok/s/chip.
Throughput at 89 tok/s/user on Muse Spark Open 816B: GB300 NVL72 hits 21064 tok/s/chip, H100 hits 1316. Per-million costs land at $0.03 and $0.25 respectively. GB300 NVL72 is 711% cheaper per token; GB300 NVL72 delivers 1501% more tok/s/chip.
GB300 NVL72 / H100 on Muse Spark Open 816B at 128 tok/s/user: 18690 / 1316 tok/s/chip, $0.03 / $0.25 per million tokens. GB300 NVL72 is 619% cheaper per token; GB300 NVL72 delivers 1320% more tok/s/chip. (Numbers reflect the default agentic-traces Β· fp4 selection for this URL β table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/chip) | GB300 NVL72:22642.0H100:1315.8 | GB300 NVL72:21063.6H100:1315.8 | GB300 NVL72:18690.2H100:1315.8 |
| Cost ($/M tok) | GB300 NVL72:$0.028H100:$0.247 | GB300 NVL72:$0.030H100:$0.247 | GB300 NVL72:$0.034H100:$0.247 |
| tok/s/MW | GB300 NVL72:10680176H100:960474 | GB300 NVL72:9935640H100:960474 | GB300 NVL72:8816145H100:960474 |
| Concurrency | GB300 NVL72:~20H100:~1 | GB300 NVL72:~17H100:~1 | GB300 NVL72:~14H100:~1 |
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.
