Qwen 3.5 397B-A17B — B300 vs GB200 NVL72 Performance per Dollar
Cost per million tokens of B300 (NVIDIA Blackwell) versus GB200 NVL72 (NVIDIA Blackwell) on Qwen 3.5 397B-A17B. Owning-hyperscaler TCO normalized by output tokens — performance per dollar across LLM workloads. Pick the more cost-efficient SKU at every target interactivity level. Use the chart controls below to switch sequences, precisions, and metrics — same interactions as the main inference chart.
On Qwen 3.5 397B-A17B at 65 tok/s/user, the per-million math comes out to $0.21 for B300 and $0.52 for GB200 NVL72; B300 delivers 148% more output per dollar.
At 102 tok/s/user on Qwen 3.5 397B-A17B, B300 costs $0.39 per million tokens; GB200 NVL72 costs $1.32. B300 is 237% more cost-efficient at this operating point.
B300 edges GB200 NVL72 at 138 tok/s/user on Qwen 3.5 397B-A17B — $0.59 per million tokens versus $3.06, a 417% cost-per-token gap. (Numbers reflect the default 1k/1k · fp8 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
GPU pricing (owning hyperscaler): B300 $2.34/GPU/hr · GB200 NVL72 $2.21/GPU/hr. Source: SemiAnalysis Market August 2025 Pricing Surveys & AI Cloud TCO Model.
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Dollar per Million Tokens | B300:$0.211GB200 NVL72:$0.522 | B300:$0.393GB200 NVL72:$1.323 | B300:$0.592GB200 NVL72:$3.057 |
| Concurrency | B300:~96GB200 NVL72:~100 | B300:~34GB200 NVL72:~19 | B300:~17GB200 NVL72:~6 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.