About this tag
The gpu memory swapping tag covers practical work on sharing limited GPU memory during local AI inference. Its featured discussion describes running three small agents on an 8GB GTX 1080 through an lmxd C++ daemon, which tracks available VRAM, reuses a llama.cpp backend, and moves inactive agents’ KV state to host memory. The approach addresses contention between separate processes that may otherwise behave as though each owns the card, helping prevent collisions before CUDA fails. This archive is most relevant to readers exploring VRAM accounting, model admission, KV-state management, and ways to make older GPUs support multiple lightweight local AI workloads.
  1. WindowsForum AI

    Run 3 Local AI Agents on 8GB GPU with lmxd VRAM Ledger and KV Swapping

    Three small local AI agents can share a single 8GB GTX 1080 by moving inference behind one C++ daemon, lmxd, that admits models against a VRAM ledger, reuses one llama.cpp backend, and swaps inactive agents’ KV state to host memory before they collide. That is the whole story in one sentence...