Multi-GPU inference
Model fit, quantization, and separate workers across NVIDIA and Intel. Count the memory. Explain the limits.
Field notes from a real homelab
Mortch
The AI Guy.
Practical local AI & homelab guides.
Real GPUs, no cloud required.
I’m Mitch. I build, run, and write about AI on the hardware in my home. Here’s what fits, what works, and what needs another try.
The home labLOCAL INFERENCE
Meet Mortch.Local AI.
Real hardware.
Human field notes.
Real hardware.
Human field notes.
5090 BeastRTX 5090
OG3090RTX 3090
14600K2 × Arc Pro B70
THREE MACHINES. FOUR SEPARATE GPUS.
Local. Inference at home
Practical. Commands & context
Human. Written by Mitch
What you’ll find hereMore from the
More from the
hardware you own.
Clear guides for homelab builders who want to get past the first model download.
The software stack
Ollama, Open WebUI, vLLM, llama.cpp, and ComfyUI. A useful front door and a clear job for each backend.
Hardware buying guides
Start with the workload. Look at VRAM, compatibility, power, and what a new card would actually add.
Guide 01 / Local AILaunch guide · Published
Four GPUs, Zero Cloud: Running LLMs Across a Heterogeneous Home GPU Cluster
Three machines, two GPU vendors, and one front door. The first guide covers model placement, honest VRAM math, and a serving setup with private access.
Read the first guide