You can’t merge physical RAM across PCs. But you can pool their memory for local AI: llama.cpp’s RPC backend splits a model’s layers across machines over your LAN, so 3 old PCs with 16 GB each can run a 32B–70B model none could load alone.

Whether you have two old laptops, a mixed Mac and Windows workstation, or spare office PCs, RAMDeck pools their memory over LAN into one cohesive AI engine.

Live benchmark results running 32B, 30B MoE, and 120B models across an 8-year-old laptop and budget desktop. Real generation speeds, logs, and token-per-second metrics.