Local AI Hardware Planner
Estimate model memory and judge feasibility on your GPU and RAM.
How it’s calculated
Weights = parameters × bits ÷ 8. KV cache ≈ 2 × layers × hidden × context × 2 bytes. Adds fixed overhead; 8 GB of RAM is reserved for the OS.
Limitations & assumptions
Effective bits are approximate. Grouped-query attention needs less cache than shown. Parameter count alone never guarantees a model runs.
FAQ
GPU or CPU?
Fully on GPU is far faster; offloading to RAM works but slows generation.