How we pick hardware
Short answer: we start from each model's published memory size, add headroom, pick the smallest tier that fits, and list new-retail hardware from Amazon, Newegg and Minisforum with the time we checked each price. FitMyModel is an informational site: we do not test, build or photograph hardware ourselves.
Where the information comes from
Pages are compiled from cited model cards, official install documentation, maker pages and store listings. There are no star ratings and no test claims.
How a model gets a tier
- Take the download size from the Ollama library at Q4_K_M, or the format the model ships in. For Whisper and SDXL we use the makers' own figures. For FLUX we use a third-party estimate, marked as one.
- Add roughly 1–4 GB for context and runtime. This is our estimate.
- Pick the smallest tier that holds it. "Tight" means it fits at short context but may spill into system RAM.
How install steps are written
The "Install and run" sections in each guide are taken from the official documentation (Ollama, LM Studio, llama.cpp, ComfyUI, whisper.cpp, NVIDIA, AMD ROCm and Intel). Each section lists the docs it used and when we checked them. We don't add flags or commands the docs don't show.
How hardware gets listed
- New condition only, from Amazon, Newegg or Minisforum. No used, refurbished or marketplace listings.
- Every price shows the store and the date and time we checked it. If we didn't check, we say "price not checked".
- Out-of-stock and third-party-seller listings are flagged, not hidden.
- We never invent benchmarks or speeds. Maker claims are quoted and attributed.