What hardware do I need to run PatentForgeLocal?

[scottconverse]

Minimum:

• 32 GB RAM
• Any modern 4+ core CPU
• 50 GB free disk space (for Ollama + model weights)

Recommended:

• 64 GB RAM
• AMD Ryzen with Radeon iGPU (780M or better) or NVIDIA GPU
• SSD storage

The default model (Gemma 4 26B) needs about 18 GB for weights. With 32 GB RAM it fits but it is tight -- you will get better performance with 48-64 GB.

GPU acceleration makes a huge difference. With a properly configured AMD Radeon 780M iGPU, expect ~8-15 tokens/second. Without GPU, CPU-only inference runs at ~3-8 tokens/second -- still usable but noticeably slower.

NVIDIA GPUs work automatically with Ollama. AMD iGPUs may need ROCm configuration on Linux (Ollama handles this automatically on Windows via DirectML). PatentForgeLocal v0.1.0 includes hardware detection scripts that configure this automatically.