Llama 3.3 70B — B200 vs H200 Performance per Dollar
Cost per million tokens of B200 (NVIDIA Blackwell) versus H200 (NVIDIA Hopper) on Llama 3.3 70B. 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.
Near the low end of the 34–159 tok/s/user interactivity band — at 65 tok/s/user — B200 runs $0.09 per million tokens on Llama 3.3 70B while H200 runs $0.14. B200 is the cheaper choice by 54%.
On Llama 3.3 70B at 97 tok/s/user, the per-million math comes out to $0.15 for B200 and $0.27 for H200; B200 delivers 81% more output per dollar.
At 128 tok/s/user on Llama 3.3 70B, B200 costs $0.29 per million tokens; H200 costs $0.65. B200 is 123% more cost-efficient at this operating point. (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.)
Chip pricing (owning hyperscaler): B200 $1.73/chip/hr · H200 $1.22/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 | B200:$0.090H200:$0.138 | B200:$0.146H200:$0.265 | B200:$0.291H200:$0.650 |
| Concurrency | B200:~128H200:~64 | B200:~74H200:~27 | B200:~27H200:~16 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.