LLM VRAM Calculator

Estimate local LLM model memory requirements by model and quantization, then compare them with your GPU VRAM.

Use this calculator when you already have a model in mind. Use the PC checker when you want to start from your hardware.

Model

Quantization / variant

Available VRAM

Selected model

Llama 3.2 3B

Q4_K_M

Provider: Meta
Parameters: 3B
Model file size: 1.7 GB
Minimum load: 2.7 GB
Recommended planning memory: 3.7 GB

Choose an available VRAM amount to compare this model variant with your GPU memory.

This variant supports GPU offload paths, so CPU/system-memory offload may be possible in compatible runtimes. AIPCFit does not count offload as a full GPU-memory fit.

Practical memory tier links

8GB VRAM guide 12GB VRAM guide

These are planning tiers, not official model requirements.

How the calculation works

AIPCFit compares catalog memory signals, not guessed benchmark data.

Model file size is a storage and planning signal, but it is not identical to runtime VRAM use.

Minimum-load planning memory is the conservative threshold AIPCFit uses before calling a model a basic GPU-memory fit.

Recommended planning memory adds headroom for normal runtime overhead and more comfortable operation.

Context length, runtime implementation, KV cache, batching, and other settings can increase memory use.

GPU offload can help compatible runtimes use CPU/system memory, but offloading is not the same as a full GPU fit.

VRAM guide links

Compare practical memory tiers.

These guides are planning tiers for local model memory, not official requirements for every model or runtime.

8GB VRAM guide 12GB VRAM guide 16GB VRAM guide 24GB VRAM guide 32GB VRAM guide 48GB VRAM guide

Related tools

Pick the workflow that matches your starting point.

What LLM can I run locally?

Use the PC checker when you are starting from your hardware and want to see which models fit.

Local AI model catalog

Browse the catalog that supplies the calculator model and quantization options.

Compatibility hub

Check GPU and model compatibility pages for specific hardware and model pairings.

Best local LLM

Compare practical model choices by hardware tier and use case.

Best local LLM for coding

Review coding-focused local model recommendations.

Methodology and limitations

Treat the result as planning guidance.

The calculator uses model variant fields from AIPCFit's catalog: estimated file size, minimum-load planning memory, recommended planning memory, quantization, and GPU-offload support. It does not infer system RAM, speed, quality, or exact context-memory growth when those values are not in the catalog.

Read the methodology