Llama 3.3 70B — H200 vs MI325X Performance per Dollar
Cost per million tokens of H200 (NVIDIA Hopper) versus MI325X (AMD CDNA 3) 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.
Push Llama 3.3 70B to 51 tok/s/user and H200 lands at $0.11 per million tokens against MI325X's $0.20 — H200 pulls ahead by 85%.
H200: $0.16 per million tokens. MI325X: $0.36. Both at 75 tok/s/user on Llama 3.3 70B, with H200 121% cheaper.
Toward the upper edge of the 27–124 tok/s/user interactivity band — at 100 tok/s/user — H200 runs $0.29 per million tokens on Llama 3.3 70B while MI325X runs $0.70. H200 is the cheaper choice by 143%. (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): H200 $1.22/chip/hr · MI325X $1.10/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 | H200:$0.107MI325X:$0.198 | H200:$0.164MI325X:$0.363 | H200:$0.287MI325X:$0.699 |
| Concurrency | H200:~64MI325X:~64 | H200:~63MI325X:~53 | H200:~24MI325X:~18 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.