Head-to-head AI inference benchmark comparison of B300 (NVIDIA Blackwell) and GB200 NVL72 (NVIDIA Blackwell) on DeepSeek V4.1 Flash 552B. 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 DeepSeek V4.1 Flash 552B at 159 tok/s/user: 123405 / 94869 tok/s/chip, $0.01 / $0.01 per million tokens. B300 is 7% cheaper per token; B300 delivers 30% more tok/s/chip.
Around the middle of the 41β515 tok/s/user interactivity band, at 278 tok/s/user on DeepSeek V4.1 Flash 552B: B300 runs 43192 tok/s/chip at $0.01/M tokens, GB200 NVL72 runs 17240 at $0.03/M. B300 is 106% cheaper per token; B300 delivers 151% more tok/s/chip.
Setting 397 tok/s/user as the target on DeepSeek V4.1 Flash 552B, B300 produces 14133 tok/s/chip ($0.04 per million tokens) and GB200 NVL72 produces 16012 ($0.03). GB200 NVL72 is 38% cheaper per token; GB200 NVL72 delivers 13% more tok/s/chip. (Numbers reflect the default agentic-traces Β· fp4 selection for this URL β table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/chip) | B300:123404.6GB200 NVL72:94868.7 | B300:43192.4GB200 NVL72:17239.8 | B300:14132.7GB200 NVL72:16012.5 |
| Cost ($/M tok) | B300:$0.005GB200 NVL72:$0.005 | B300:$0.015GB200 NVL72:$0.030 | B300:$0.044GB200 NVL72:$0.032 |
| tok/s/MW | B300:64949780GB200 NVL72:50731926 | B300:22732826GB200 NVL72:9219140 | B300:7438287GB200 NVL72:8562823 |
| Concurrency | B300:~35GB200 NVL72:~50 | B300:~12GB200 NVL72:~4 | B300:~4GB200 NVL72:~8 |
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.
