← All Setups
home-office
From $800
AI Developer Workstation on a Budget
Best for: ML engineers, AI researchers, and developers experimenting with local LLMs
Setup Steps
1
Maximise RAM first
Local LLMs load entirely into RAM or VRAM. A 7B model needs 8GB, a 13B needs 16GB, and a 70B needs 48GB+ with quantisation. 32GB RAM is the minimum useful configuration for AI work.
2
Choose a machine with GPU or buy an eGPU
An RTX 4060 or RTX 3080 GPU enables CUDA acceleration for inference and fine-tuning. Apple Silicon (M3 Pro or M4 Pro) uses unified memory and outperforms most consumer GPUs per watt.
3
Set up Ollama for local model serving
Install Ollama, pull Llama 3.1 8B and Mistral 7B as baseline models. Use Open WebUI as the front end. The whole stack is free and runs on any machine with 16GB RAM.
4
Allocate a fast SSD for model storage
Models are 4–40GB each. Loading from a slow HDD defeats the purpose of local inference. A Samsung 990 Pro NVMe gives 7,450MB/s reads — models load in under 10 seconds.