DeepSeekv4 Pro 0813 1.6T — H200 vs MI300X Performance per Dollar
Cost per million tokens of H200 (NVIDIA Hopper) versus MI300X (AMD CDNA 3) on DeepSeekv4 Pro 0813 1.6T. 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.
On DeepSeekv4 Pro 0813 1.6T at 13 tok/s/user, the per-million math comes out to $0.92 for H200 and $1.01 for MI300X; H200 delivers 10% more output per dollar.
At 19 tok/s/user on DeepSeekv4 Pro 0813 1.6T, H200 costs $0.98 per million tokens; MI300X costs $2.03. H200 is 106% more cost-efficient at this operating point.
H200 edges MI300X at 26 tok/s/user on DeepSeekv4 Pro 0813 1.6T — $1.09 per million tokens versus $3.26, a 199% cost-per-token gap. (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 · 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 | H200:$0.915MI300X:$1.005 | H200:$0.985MI300X:$2.030 | H200:$1.089MI300X:$3.255 |
| Concurrency | H200:~117MI300X:~84 | H200:~79MI300X:~29 | H200:~52MI300X:~13 |
Cost per Million Total Tokens (Owning - Hyperscaler) vs. Interactivity
DeepSeek V4 Pro 0813 1.6T • FP4 • 8K / 1K • Source: SemiAnalysis InferenceX™
DeepSeek V4 Pro 0813 1.6T • FP4 • 8K / 1K • Source: SemiAnalysis InferenceX™TCO $/chip/hr: VR200: 3.61H100: 1.17H200: 1.22B200: 1.73B300: 2.26GB200: 1.86GB300: 2.31MI300X: 0.95MI325X: 1.1MI355X: 1.5RTX6000PRO: 0.68Jalapeño (Teacup): 1.47
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
Interactivity is the rate at which a single user receives generated tokens while the model streams its answer — how quickly new words appear on screen. Higher values feel snappier; operators trade it against batch throughput.
Estimated infrastructure cost of producing one million total tokens (input + output) at this operating point, priced with the all-in hourly ownership cost of a hyperscaler operator. Lower is cheaper.
Formula: $/Mtok = all-in cost per chip-hour ($) × 1,000,000 ÷ (3,600 × total tok/s/chip)