Qwen 3.5 397B-A17B — GB200 NVL72 vs MI355X Performance per Dollar
Cost per million tokens of GB200 NVL72 (NVIDIA Blackwell) versus MI355X (AMD CDNA 4) 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 edges MI355X at 90 tok/s/user on Qwen 3.5 397B-A17B — $0.05 per million tokens versus $0.10, a 107% cost-per-token gap.
Push Qwen 3.5 397B-A17B to 134 tok/s/user and GB200 NVL72 lands at $0.08 per million tokens against MI355X's $0.14 — GB200 NVL72 pulls ahead by 74%.
GB200 NVL72: $0.14 per million tokens. MI355X: $0.20. Both at 178 tok/s/user on Qwen 3.5 397B-A17B, with GB200 NVL72 36% cheaper. (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 · MI355X $1.50/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:$0.050MI355X:$0.103 | GB200 NVL72:$0.082MI355X:$0.143 | GB200 NVL72:$0.143MI355X:$0.195 |
| Concurrency | GB200 NVL72:~1070MI355X:~22 | GB200 NVL72:~279MI355X:~11 | GB200 NVL72:~45MI355X:~6 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.