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.
Only GB300 NVL72 has cost data at 169 tok/s/user on Qwen 3.5 397B-A17B — $0.04 per million tokens. MI300X is unmeasured at this target.
GB300 NVL72 costs $0.08 per million tokens at 315 tok/s/user on Qwen 3.5 397B-A17B; we have no MI300X benchmark data at this exact target.
At 461 tok/s/user on Qwen 3.5 397B-A17B, GB300 NVL72 comes in at $0.23 per million tokens. MI300X hasn't been benchmarked at this operating point. (Numbers reflect the default 8k/1k · fp4 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
Chip pricing (owning hyperscaler): GB300 NVL72 $2.31/chip/hr · MI300X $0.95/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 | GB300 NVL72:$0.037MI300X:— | GB300 NVL72:$0.076MI300X:— | GB300 NVL72:$0.235MI300X:— |
| Concurrency | GB300 NVL72:~417MI300X:— | GB300 NVL72:~120MI300X:— | GB300 NVL72:~16MI300X:— |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.