Llama 3.3 70B — H200 vs MI325X
Head-to-head AI inference benchmark comparison of H200 (NVIDIA Hopper) and MI325X (AMD CDNA 3) on Llama 3.3 70B. Latency, throughput, and cost across LLM workloads. Use the chart controls below to switch sequences, precisions, and metrics — same interactions as the main inference chart.
Setting 34 tok/s/user as the target on Llama 3.3 70B, H200 produces 3521 tok/s/chip ($0.10 per million tokens) and MI325X produces 1769 ($0.17). H200 is 79% cheaper per token; H200 delivers 99% more tok/s/chip.
At 58 tok/s/user interactivity on Llama 3.3 70B, H200 delivers 2516 tok/s/chip at $0.13 per million tokens; MI325X delivers 1354 tok/s/chip at $0.23. H200 is 68% cheaper per token; H200 delivers 86% more tok/s/chip at this point.
H200 posts 1705 tok/s/chip for $0.20 per million tokens at 83 tok/s/user on Llama 3.3 70B; MI325X posts 756 tok/s/chip for $0.40. H200 is 103% cheaper per token; H200 delivers 125% more tok/s/chip. (Numbers reflect the default 8k/1k · fp8 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/chip) | H200:3520.9MI325X:1769.3 | H200:2516.2MI325X:1354.4 | H200:1704.5MI325X:756.5 |
| Cost ($/M tok) | H200:$0.096MI325X:$0.173 | H200:$0.135MI325X:$0.226 | H200:$0.199MI325X:$0.404 |
| tok/s/MW | H200:2570005MI325X:1046907 | H200:1836618MI325X:801393 | H200:1244164MI325X:447630 |
| Concurrency | H200:~32MI325X:~20 | H200:~23MI325X:~25 | H200:~9MI325X:~9 |
Inference Performance
Agentic inference metrics from the AgentX scenario and fixed-sequence inference metrics across models, hardware configurations, and serving parameters.
Total Tokens per $1 USD (Owning - Hyperscaler) vs. Interactivity
Llama 3.3 70B Instruct • FP4 • 8K / 1K • Source: SemiAnalysis InferenceX™
TCO $/chip/hr: H100: 1.17H200: 1.22B200: 1.73B300: 2.26GB200: 1.86GB300: 2.31MI300X: 0.95MI325X: 1.1MI355X: 1.5RTX6000PRO: 0.68
Source: SemiAnalysis Market July 2026 Pricing Surveys & AI Cloud TCO Model
Note: Disaggregated inference configurations (e.g., MoRI SGLang, Dynamo TRTLLM) calculate cost per million tokens per decode chip or per prefill chip, rather than per total chip count. This makes direct token cost comparison with aggregated configs not an apples-to-apples comparison.
Note: Disaggregated inference configurations (e.g., MoRI SGLang, Dynamo TRTLLM) calculate input throughput per decode chip or per prefill chip, rather than per total chip count. This makes direct input throughput comparison with aggregated configs not an apples-to-apples comparison.
Note: Disaggregated inference configurations (e.g., MoRI SGLang, Dynamo TRTLLM) calculate output throughput per decode chip or per prefill chip, rather than per total chip count. This makes direct output throughput comparison with aggregated configs not an apples-to-apples comparison.
Note: Disaggregated inference configurations (e.g., MoRI SGLang, Dynamo TRTLLM) calculate power per decode chip or per prefill chip, rather than per total chip count. This makes direct power comparison with aggregated configs not an apples-to-apples comparison.
Note: Disaggregated inference configurations (e.g., MoRI SGLang, Dynamo TRTLLM) calculate Joules per decode chip or per prefill chip, rather than per total chip count. This makes direct Joules per token comparison with aggregated configs not an apples-to-apples comparison.
No data available
Please change the model, sequence, precision, date range or chip selection.
Shift+Scroll to zoom • Drag to pan • Double-click to reset • Click a point to pin tooltip