What can my GPU run?
The VRAM calculator answers "will this model fit?" one model at a time. This page works the other way round: tell it what hardware you have and it checks every model and workload we track at once, then shows the highest-quality quantization that still fits.
What can my GPU run?
Pick your graphics card or Mac, or let your browser detect it. Every model and workload we track is checked at once.
Same numbers and 90% rule as the VRAM calculator and VRAM table. GPU detection runs only in your browser and nothing is sent anywhere. Checked 2026-10-23.
Browse by VRAM size
No JavaScript, or just want a quick list? Each page below shows what fits in that much VRAM or Mac unified memory with an 8K-token conversation.
8 GB12 GB16 GB24 GB32 GB48 GB64 GB96 GB128 GB
How we decide what fits
- Language models: we add the real Hugging Face download size at each precision to the FP16 KV cache for your context length. A model fits when that total stays under 90% of your memory. We start at 16-bit and step down to 8-bit, then 4-bit, and show the first one that fits.
- More than one GPU: we add the cards' memory together, because llama.cpp, Ollama and vLLM can split a model's layers across GPUs. Image generation, video editing and games mostly use one card, so those are checked against a single card.
- Image, video and gaming: there's no formula for these, so we use the published minimum and comfortable ranges listed in the calculator, with sources.
Speed (tokens per second) isn't shown. It depends on memory bandwidth, the runtime and the settings, and we only publish numbers we can source.
Download the data: models (CSV) · models (JSON) · GPU prices (JSON). Every row carries its source links and checked date.