Llama 3.3 70B — H100 vs MI355X Performance per Dollar
Cost per million tokens of H100 (NVIDIA Hopper) versus MI355X (AMD CDNA 4) 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.
At 53 tok/s/user on Llama 3.3 70B, H100 costs $0.22 per million tokens; MI355X costs $0.19. MI355X is 20% more cost-efficient at this operating point.
MI355X edges H100 at 72 tok/s/user on Llama 3.3 70B — $0.26 per million tokens versus $0.42, a 65% cost-per-token gap.
Push Llama 3.3 70B to 91 tok/s/user and H100 lands at $1.07 per million tokens against MI355X's $0.44 — MI355X pulls ahead by 145%. (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): H100 $1.17/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 | H100:$0.222MI355X:$0.185 | H100:$0.421MI355X:$0.255 | H100:$1.070MI355X:$0.438 |
| Concurrency | H100:~64MI355X:~47 | H100:~50MI355X:~53 | H100:~14MI355X:~22 |
Inference Performance
Agentic inference metrics from the AgentX scenario and fixed-sequence inference metrics across models, hardware configurations, and serving parameters.
Cost per Million Total Tokens (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