HARDWARE CHECK

Can my PC run local AI?

Usually the answer is “yes, but model choice matters”. A modest computer can run small models; larger models need more memory and may become impractically slow if too much work falls back to the CPU.

1. Find your actual available memory

On a PC, check dedicated GPU VRAM and system RAM separately. On Apple Silicon, CPU and GPU share unified memory. Use our VRAM and RAM checking guide.

2. Choose a model size and weight format

Quantised weights are much smaller than FP16 weights. A model download still needs extra working memory for context and runtime overhead.

3. Start with a modest context

Long context windows can add substantial cache memory. Start short and increase later.

4. Use the calculator

The HostingLinks memory calculator estimates GPU and system memory under consistent assumptions.

5. Expect CPU fallback to change speed

CPU offload may let an oversized model run, but generation can be much slower than a model that fits cleanly in GPU memory.

Do not buy hardware before proving the workflow

Run one small model first, then use the upgrade comparison to see what more memory would unlock.

Next: how much VRAM do you need? or run your first Ollama model.