GPUFits
Beginner Learning Path
A 6-step learning path for complete beginners: understand VRAM and quantization, pick the right hardware, install a runtime, and run your first local model — with a hands-on task at every step.
Who this is for: complete beginners to local deployment — no idea what VRAM is, never heard of quantization, no runtime installed. You can follow along.
What you'll be able to do: understand the basics of VRAM and quantization, use our tools to judge what your hardware can run, and run your first local LLM on your own machine.
- 1
What Is Local Deployment? Local LLMs vs Cloud APIs in Plain Terms
Local deployment in plain terms: the model runs on your own machine — data stays put, works offline, heavy use gets cheaper. Plus when an API wins.
Start this step →
- 2
Hardware 101: VRAM Decides If It Runs, Bandwidth Decides How Fast
GPU specs for beginners: VRAM is desk size (does the model fit?); bandwidth is hand speed (how fast text comes out). With real numbers.
Start this step →
- 3
Choosing a Model and Quant: B, GGUF, and Q4_K_M Explained
A zero-background guide to model naming: what the B means, what a GGUF file is, how to choose between Q4_K_M and Q8_0, and where to download on Hugging Face.
Start this step →
- 4
Your First Run: The 10-Minute Ollama Route
A follow-along guide for total beginners: install Ollama, run Llama 3.2 3B with one command, have your first chat, and check usage with ollama ps.
Start this step →
- 5
Too Slow or Won't Fit? A Five-Step Checklist
Model won't fit or runs at a crawl? Troubleshoot in a fixed order: lower the quant, shorten context, enable q8 KV cache, CPU offload, then multi-GPU.
Start this step →
- 6
The Roadmap: Where to Go After Your First Model
Four next steps after your first model: the VRAM cost of long context, how MoE changes the math, Ollama vs llama.cpp vs vLLM, and when to rent cloud GPUs.
Start this step →