Behind FitMyLLM.
Who I am
I'm an MLOps engineer. My work is getting machine learning models to run on real hardware, at a speed people can actually work with. It's the same problem this site solves, for your machine instead of mine.
Why this exists
Open models have got good enough that a lot of everyday work runs on hardware people already own. Whether that is worth doing depends on what you need: keeping data on your machine, avoiding a per-token bill, working without a connection, or simply wanting to see how far your card goes.
The awkward part is knowing what your machine can actually run. It depends on VRAM, memory bandwidth, quantisation, context length and how those interact, and getting it wrong is easy in both directions: a model too big crawls, a model too small wastes hardware you paid for.
FitMyLLM answers that one question. Tell it your hardware, it tells you what runs and how fast. Where the answer is “nothing you want to run fits”, it says that too.
My goal
I want anyone to be able to choose, install, and experiment with a local AI model, whatever their technical background. Not just developers.
If you've never opened a terminal, you should still be able to land here, see what your computer can run, and be up and running in five minutes. That's the bar I'm building towards.
How I approach it
Running a model on your own machine is useful when privacy, cost or working offline matter to you. It is not a better way to live, and this site is not here to talk you into it.
A frontier model you rent will beat anything your GPU runs, and for occasional use it is cheaper too. If that is your situation, the honest recommendation is to use it.
I don't promote any vendor. I don't inflate benchmarks. If a $800 used GPU performs as well as a $1600 new one for your use case, I'll tell you.
My estimates are approximations. Real-world benchmarks from actual users make recommendations more accurate for everyone.
Limitations
My performance estimates are based on calibrated formulas, not measured results. Real-world speed depends on your system, your inference engine, your settings, and a hundred other variables. See the Methodology page for full details.
Independence & transparency
FitMyLLM is an independent project. I have no business relationship with Ollama, HuggingFace, NVIDIA, AMD, Apple, or any model provider. Recommendations come from data and benchmarks, nothing else.
Some hardware links are affiliate links. If you purchase through them, I may earn a small commission at no extra cost to you. The same algorithm ranks hardware regardless of whether an affiliate link exists.
FitMyLLM is free and always will be. If it helped you pick the right model or avoid buying the wrong GPU, consider supporting development.
Questions, corrections, partnerships, or a measurement that looks wrong: fitmyllm@gmail.com. Nothing on this site is so settled that a good correction is unwelcome.