Head-to-head AI inference benchmark comparison of B300 (NVIDIA Blackwell) and GB200 NVL72 (NVIDIA Blackwell) on GLM 5.3 744B. 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
B300 / GB200 NVL72 on GLM 5.3 744B at 137 tok/s/user: 14617 / 14105 tok/s/chip, $0.04 / $0.04 per million tokens. GB200 NVL72 is 17% cheaper per token; B300 delivers 4% more tok/s/chip.
Around the middle of the 81β303 tok/s/user interactivity band, at 192 tok/s/user on GLM 5.3 744B: B300 runs 7494 tok/s/chip at $0.08/M tokens, GB200 NVL72 runs 8833 at $0.06/M. GB200 NVL72 is 43% cheaper per token; GB200 NVL72 delivers 18% more tok/s/chip.
Setting 248 tok/s/user as the target on GLM 5.3 744B, B300 produces 6148 tok/s/chip ($0.10 per million tokens) and GB200 NVL72 produces 4737 ($0.11). B300 is 7% cheaper per token; B300 delivers 30% more tok/s/chip. (Numbers reflect the default agentic-traces Β· fp4 selection for this URL β table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/chip) | B300:14616.5GB200 NVL72:14104.6 | B300:7494.0GB200 NVL72:8833.2 | B300:6148.4GB200 NVL72:4737.1 |
| Cost ($/M tok) | B300:$0.043GB200 NVL72:$0.037 | B300:$0.084GB200 NVL72:$0.058 | B300:$0.102GB200 NVL72:$0.109 |
| tok/s/MW | B300:7692921GB200 NVL72:7542557 | B300:3944228GB200 NVL72:4723651 | B300:3236002GB200 NVL72:2533187 |
| Concurrency | B300:~12GB200 NVL72:~74 | B300:~10GB200 NVL72:~45 | B300:~2GB200 NVL72:~5 |
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.
