Can your PC run
it locally?
Drop your DxDiag file. See which quantizations fit your VRAM and how fast they will run. No install, no guessing.
Check which AI models your GPU can run
Idxdiag tells you which local AI models your computer can actually run, and how fast. Search a model like Llama 3.1 or Stable Diffusion, share your system details, and see which quantization levels fit your graphics memory along with a rough speed estimate. Everything is scored against your real hardware, so you know before you download.
Model memory is computed, never guessed. The core idea is simple. The video memory a model needs is roughly its number of parameters multiplied by the bytes used per parameter, plus overhead for the working state during generation. A smaller quantization uses fewer bytes per parameter, which is why the same model can fit a smaller card once it is compressed. Idxdiag shows the memory each quantization needs and marks the ones that fit your card.
Mixture of experts models are counted by their total parameters, not just the parameters active on each token. Even when only a portion of the network runs at once, the whole model still has to sit in memory, so a mixture of experts model with a very large total will not fit a small card even though it is efficient to run. When a model is too large for any consumer card, we say so plainly instead of showing a green light you cannot use.
Popular right now
Browse allWhy Idxdiag?
Recently added Live
Browse allBrowse by category
What we do best
Idxdiag reads your graphics memory straight from a DxDiag report, so you never need to look up your own card. It shows every quantization of a model and marks exactly which ones fit your video memory, each with a speed estimate. It counts memory honestly, including the overhead that generation needs and the full size of mixture of experts models. And it checks games and creative software the same way, all against the hardware you already own.
Popular hardware questions
Frequently asked questions
What GPU do I need for Llama 70B?
A 70 billion parameter model needs a large amount of video memory, often more than a single consumer card offers unless it is quantized. Idxdiag shows the memory each quantization of Llama 70B needs and marks which ones fit your graphics card.
How much VRAM does Stable Diffusion need?
Stable Diffusion is far smaller than a large language model and runs comfortably on many mid range cards. Search the model, share your graphics memory, and Idxdiag confirms whether your card has enough headroom.
Can I run a 7B model on 8GB VRAM?
Often yes, once the model is quantized. A 7 billion parameter model at a four bit quantization fits many 8GB cards with room for the working state. Idxdiag shows the exact memory each quantization needs so you can be sure.
What is quantization?
Quantization stores a model’s numbers with fewer bits, which shrinks how much memory it needs while keeping most of its quality. A four bit version of a model uses far less video memory than the full precision version, which is what lets larger models fit smaller cards.
How do you calculate the memory a model needs?
We take the number of parameters, multiply by the bytes used per parameter for the chosen quantization, and add overhead for the working state during generation. For mixture of experts models we count the total parameters, because the whole model must sit in memory even when only part of it runs.
Do you store my system file?
No. Your dxdiag report is read in memory to read your graphics memory and other specs, then discarded. Nothing is kept and no account is required.
See it in action
Click below to see a live verdict. We'll load a sample mid-range gaming PC and check it against Cyberpunk 2077's requirements.
