AI Hardware Store
Browse Mini PCs, NAS, workstations, servers and edge AI hardware — every system can be paired with installation service and software innovation by ⏻ On Intelligence.
| Photo | Product | Price | Capacity |
|---|
Understanding model capacity
Model capacity refers to the number of parameters an AI model contains (e.g. 70B = 70 billion parameters). Larger models generally require more memory and computing power, but can offer improved reasoning and output quality.
Precision (FP16 / FP8 / FP4) affects how much memory a given model capacity requires: lower-precision formats roughly halve memory requirements per step down (FP16 → FP8 → FP4), at some cost to output precision.
| Model size | Typical memory (FP16) | Typical memory (FP4) |
|---|---|---|
| ~70B | ~140GB | ~40GB |
| ~120B | ~240GB | ~65GB |
| ~405B | ~810GB | ~220GB |
| ~700B | ~1.4TB | ~380GB |