Qwen 3.5 397B-A17B — GB300 NVL72 vs MI300X Performance per Dollar
Cost per million tokens of GB300 NVL72 (NVIDIA Blackwell) versus MI300X (AMD CDNA 3) 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.
Push Qwen 3.5 397B-A17B to 43 tok/s/user and GB300 NVL72 lands at $0.11 per million tokens against MI300X's $0.89 — GB300 NVL72 pulls ahead by 701%.
GB300 NVL72: $0.24 per million tokens. MI300X: $1.63. Both at 53 tok/s/user on Qwen 3.5 397B-A17B, with GB300 NVL72 579% cheaper.
Toward the upper edge of the 34–71 tok/s/user interactivity band — at 62 tok/s/user — GB300 NVL72 runs $0.43 per million tokens on Qwen 3.5 397B-A17B while MI300X runs $2.52. GB300 NVL72 is the cheaper choice by 479%. (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): GB300 NVL72 $2.65/GPU/hr · MI300X $1.12/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 | GB300 NVL72:$0.111MI300X:$0.893 | GB300 NVL72:$0.240MI300X:$1.633 | GB300 NVL72:$0.434MI300X:$2.517 |
| Concurrency | GB300 NVL72:~2063MI300X:~34 | GB300 NVL72:~701MI300X:~15 | GB300 NVL72:~271MI300X:~8 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.