Qwen 3.5 397B-A17B — GB200 NVL72 vs H100 Performance per Dollar
Cost per million tokens of GB200 NVL72 (NVIDIA Blackwell) versus H100 (NVIDIA Hopper) 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.
GB200 NVL72: $0.05 per million tokens. H100: $0.24. Both at 70 tok/s/user on Qwen 3.5 397B-A17B, with GB200 NVL72 415% cheaper.
Around the middle of the 36–172 tok/s/user interactivity band — at 104 tok/s/user — GB200 NVL72 runs $0.07 per million tokens on Qwen 3.5 397B-A17B while H100 runs $0.30. GB200 NVL72 is the cheaper choice by 359%.
On Qwen 3.5 397B-A17B at 138 tok/s/user, the per-million math comes out to $0.13 for GB200 NVL72 and $0.37 for H100; GB200 NVL72 delivers 188% more output per dollar. (Numbers reflect the default 8k/1k · fp8 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
Chip pricing (owning hyperscaler): GB200 NVL72 $1.86/chip/hr · H100 $1.17/chip/hr. Source: SemiAnalysis Market July 2026 Pricing Surveys & AI Cloud TCO Model.
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Dollar per Million Tokens | GB200 NVL72:$0.046H100:$0.238 | GB200 NVL72:$0.066H100:$0.301 | GB200 NVL72:$0.127H100:$0.366 |
| Concurrency | GB200 NVL72:~1305H100:~20 | GB200 NVL72:~483H100:~11 | GB200 NVL72:~72H100:~9 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.