Can your PC run this AI model?
Search any AI model and instantly see the RAM and VRAM you need, quantization by quantization — using real GGUF sizes and clearly labeled estimates.
Get my detected specs
Get-CimInstance Win32_VideoController | Select Name, AdapterRAM
Find by graphics card
Most downloaded GGUF models
Qwen3-Coder-30B-A3B-Instruct-GGUF
gpt-oss-20b-GGUF
Qwen-AgentWorld-35B-A3B-GGUF
Qwen3-4B-GGUF
Meta-Llama-3.1-8B-Instruct-GGUF
Jan-v3.5-4B-gguf
Qwen3-8B-GGUF
Qwen3-0.6B-GGUF
Qwen3-14B-GGUF
Qwen3-1.7B-GGUF
Qwen3-32B-GGUF
Qwen3-30B-A3B-GGUF
Numbers you can trust
- Real GGUF file sizes read live from the model hub — no estimates.
- KV cache uses the model's architecture when available; fallback estimates are labeled.
- No hardcoded lists — model data updates automatically.
Want the details? Read exactly how we calculate every number.
How to check if your PC can run an AI model
Running a large language model locally means loading it entirely into your computer's memory — ideally your graphics card's VRAM, which is far faster than system RAM. Whether a given model fits depends on three things: its size, the quantization you pick, and how long a context you run. Here's how to find your answer in under a minute.
Search the model
Type a model name above — Llama, Qwen, Mistral, DeepSeek, Gemma and thousands more in GGUF format. You'll land on a page with its real memory needs per quantization.
Enter your hardware
Detect your GPU automatically, or pick your graphics card from the list. The tool compares each quantization against your VRAM and system RAM.
Read the verdict
Each quantization is marked fits in VRAM, runs with offload (slower), or insufficient — so you pick the best quality your machine can hold.
What decides whether a model fits?
Weights & quantization
Quantization shrinks a model by storing its weights at fewer bits. An 8-bit copy is roughly half the size of 16-bit; 4-bit halves it again, with a small quality cost. Picking the right level is usually what makes a model fit. Quantization explained →
VRAM vs system RAM
A model held entirely in GPU VRAM runs fast. If it's too big, part spills over to system RAM ("offloading") — it still works, but slows down. Nothing fitting in VRAM and RAM combined won't run at all. How much VRAM? · How much RAM?
Context length
The longer the conversation, the more memory the KV cache needs on top of the weights. Modern designs like grouped-query, hybrid and latent attention keep this cost low — which is why our figures come from each model's real architecture.
What can each GPU run?
As a rough starting point, here's the largest model size that typically fits at a balanced 4-bit quantization. Your exact result depends on the model and context — check your own card for precise numbers.
| VRAM | Typically runs (4-bit) | Example cards |
|---|---|---|
| 8 GB | up to ~8B models | RTX 3050, 4060 |
| 12 GB | up to ~13B models | RTX 3060, 4070 |
| 16 GB | up to ~22B models | RTX 4060 Ti, 5070 Ti |
| 24 GB | up to ~34B models | RTX 3090, 4090 |
| 32 GB | up to ~49B, or a 70B at low bits | RTX 5090 |
See the best model for your exact VRAM → · Browse all graphics cards →
Frequently asked questions
Do I need a graphics card to run an AI model locally?
Not strictly — small models run on CPU and system RAM alone, just slowly. A GPU with enough VRAM is what makes local models genuinely fast, because the whole model can stay in fast graphics memory. This tool shows both cases: fully in VRAM, or partly offloaded to RAM.
What is GGUF, and why does it matter?
GGUF is the file format used by llama.cpp, Ollama and LM Studio to run models locally. Each GGUF file is a specific quantization of a model, so its size on disk closely tracks the memory it needs. We read those real file sizes rather than estimating.
Which quantization should I choose?
Pick the highest-quality quantization that still fits your VRAM. Q4_K_M is a popular balance — most of the quality at roughly half the memory of 8-bit. If you have room to spare, Q5 or Q6 add a little more; if you're tight, a smaller quant still runs. Full breakdown →
How accurate are these numbers?
Weight sizes, parameter counts and KV-cache math are exact, computed from each model's published files and configuration. The only figure we label as an estimate is inference speed, which depends on your specific setup. See our methodology.
What tools do I use to actually run the model?
Once you know a model fits, Ollama and LM Studio are the easiest ways to download and run it. This site tells you what will fit; those tools do the running.