CheckerWhat fits my GPU?ModelsWhat to buyEmbeddingsData & methodSubmitAPISponsor

Can your GPU or Mac run that AI model?

Short answer: Whether a model runs on your GPU comes down to memory, not speed. Will It Run? compares the memory an open model needs at your quantization—weights, KV cache and buffers—with your VRAM or Mac unified memory. It covers 62 current models and shows measured llama.cpp speeds where they exist. Memory figures are estimates.

Data checked 2026-10-21 · memory figures are estimates; measured speeds link their source

Pick your hardware (add several GPUs if you have them), your system RAM, a model and a quantization. We estimate the memory it needs and show measured speeds from public community benchmarks where they exist. 62 current models, newest first.

New: What fits my GPU? · free API v1 · one-command benchmark (curl -fsSL https://willitrunllm.com/bench.py | python3 -)

The GB box shows each card's VRAM. Type your own number for custom VRAM, or pick ✏️ Custom VRAM.
Type a name or paste a repo (user/name). We read the real GGUF file sizes straight from Hugging Face.

Will It Run? Pro NOT BUILT YET

A personal setup for your exact hardware (Ollama / llama.cpp configs ready for your coding agent), an alert when a new model fits, and a one-click cloud fallback when it doesn't.

Not built yet. Join the waitlist, $10/mo planned, no charge now. What's planned

Email only. No payment info.

Embedding models for RAG

Open embedding models that turn text into vectors for search/RAG. The fit column checks whether each one fits in the VRAM left over after the chat model selected above.

Loading…

Get the weekly "new models & what it takes to run them" email

Free. One email a week. Unsubscribe anytime.

Newest models

All 81 models →

Best LLM for your VRAM

8 GB · 12 GB · 16 GB · 24 GB · 32 GB · 48 GB · 80 GB

FAQ

How do I know if an LLM will run on my GPU?

Compare the memory the model needs at your quantization (weights + KV cache + buffers) with your VRAM. The checker does this for 62 current models and shows measured community speeds where available.

What do the verdicts mean?

Fits = fits fully in GPU memory; tight = fits with under 10% headroom; offload = needs system RAM and runs slower; no = does not fit. All memory figures are estimates.

How much VRAM do I need to run a local LLM?

Our estimate is parameters × bits per weight ÷ 8, plus 10% and 1 GB for a ~4k context. Q4_K_M is 4.85 bits per weight and Q8_0 is 8.5, so the quantization you pick matters as much as the model size.

Can I run a local LLM on a Mac?

Yes, on Apple Silicon the GPU uses unified memory. We assume macOS lets the GPU use about 75% of memory on 64 GB+ Macs and 67% below. You can raise that limit with sysctl iogpu.wired_limit_mb, at your own risk.

Do two GPUs make a model run twice as fast?

No. VRAM is summed across GPUs for the fit verdict, so two cards can hold a bigger model. Speed is not additive.

Are the tokens-per-second numbers real?

Measured speeds are community llama-bench results, each linked to its source. For other models we estimate from the measured Llama 2 7B Q4_0 speed on the same chip, scaled by weight size, and label it as an estimate. We don't estimate speed for MoE models.