Upgrade your GPU or rent one?
Want to run a model your computer can't handle? Compare buying a card that runs it with renting a cloud GPU by the hour, based on how much you'll actually use it.
How to decide
Rent if you'll use a big model a few hours a week, want to try one before spending money, or need more memory than any consumer card has. You pay by the hour, stop any time, and the same ollama run command works on a rented machine.
Buy if you'll use it most days, want everything offline and private, or also want the card for gaming or creative work. A card that's busy many hours a week usually pays for itself within a year or two.
Do neither if a smaller model does the job. The picker finds the best model your current computer runs well; for many everyday tasks that's enough.
Card prices are US launch prices; used cards are often much cheaper. Rental prices are typical ranges on community GPU clouds and change often. Electricity isn't included.
Run it in the cloud
Try the model on a rented GPU first.
Renting for an hour is the cheapest way to find out whether a bigger model is worth buying hardware for.
Frequently asked questions
How much does it cost to rent a GPU for AI?
A 24 GB card such as an RTX 4090 typically rents for about $0.25 to $0.70 an hour on community clouds like RunPod and Vast.ai. An 80 GB A100 or H100 is roughly $1.20 to $2.70 an hour.
Is it cheaper to rent or buy a GPU for local LLMs?
For a few hours a week, renting is usually cheaper for a long time. If you use it daily, buying often pays off within one to two years, and you keep the card.
Can I run a 70B model at home?
A 70B model needs about 44 GB at Q4_K_M, more than any single consumer graphics card. Options are a Mac with 64 GB or more of unified memory, two 24 GB cards, or renting a 48 to 80 GB cloud GPU.