LLMGauge

LLMGauge · local model compatibility

Can your computer run this local LLM?

Choose a model, describe your hardware, and get a transparent first-pass compatibility assessment.

Start here

Choose the question you want to answer

Use the calculator for one model, or compare the catalog against your hardware.

Check one specific model

Pick a model and quantization when you already know what you want to run.

Use the calculator below

Discover models for your PC

Enter your hardware once and receive a deterministic list of suitable catalog models.

Open recommendations

You will need your CPU, system RAM, GPU (if you have one), dedicated VRAM, and operating system. Results are approximate planning guidance, not performance guarantees.

Browse examples from the catalog: Qwen3 4B has a model profile, and the GeForce RTX 3060 12GB page documents a dedicated-GPU profile.

01

Choose a model

Select one model and its quantization.

Quantization changes the model’s memory requirement. Higher-bit options generally preserve more quality.

Meta's compact instruction-tuned Llama model for lightweight local text generation.

02

Describe your hardware

Use the values available in your system information.

Prefilled values are starter examples, not readings detected from your device. Edit them for your PC. Scanning is optional and starts only when you choose Scan my device.

Select Apple Silicon explicitly for a single shared memory pool. macOS or a scan never enables this mode automatically.

Optional: scan your device

This stays in your browser and reads only coarse hints. It never submits or stores hardware information.

For example, Intel Core i5 or AMD Ryzen 5.
Choose no GPU for a CPU-only evaluation.
Enter dedicated GPU memory in GiB.
Total memory available to the operating system, for example 16 GiB.
Advanced settings

Optional runtime planning

These inputs can establish a cache-memory placement assumption. They do not verify drivers, backend support, actual allocation, or speed.

Choose llama.cpp when that is the runtime you plan to use.
This is an assumption, not a capability check.
The preference is advisory and does not override the result.
Optional planning target for the FP16 KV-cache estimate. Choose a common value below or enter a custom number. Cache memory is included in fit only when its placement is explicit; otherwise it remains advisory.

Estimates are approximate memory guidance, not performance benchmarks or guarantees.