A 9-machine local AI homelab: 5 agents on one dual-3090 llama-server
StandardLovers · reddit · 2026-11-23
A Redditor showcased a fully local AI homelab: 9 machines including a dual-RTX-3090 inference rig (48GB VRAM, llama.cpp with NCCL) serving five agents — two personal assistants, IDE and CLI coding agents, and a family agent — on Qwen 27B Q8 with 256k context and 2000 tok/s prefill. Highlights: a kernel module firmware-patching the 3090s' FP16/BF16 FMA issue rate (+10% at long context), Tailscale mesh, hardened firewalls, 3-2-1 backups, and 12 cron automation jobs. Everything is green; the author is soliciting new ideas.
More from coding & agent
- Apple cuts per-dev daily token budget from $300 to $100; cheaper-model strategies follow — Zachly · 2026-11-23
- Fine-tuned Gemma 4 E4B drives 100 villagers offline in a rumor-spread simulation game — AlternativeEnough352 · 2026-11-23
- A lightweight script makes CLI agents debate each other to a consensus write-up — reddit_kwr · 2026-11-23
- Nous ships official Hermes Blender plugin: agents drive live 3D scenes via MCP — Teknium · 2026-11-23
- Dev builds Battlefield 1942-style browser FPS with GPT-6.2: 16v16 battles, vehicles, AI squads — VraserX · 2026-11-23
- Dev spins up hundreds of AI agents in a livestreamed attempt to solve P = NP — amantinband · 2026-11-23