What this estimate does
It maps broad RAM and VRAM bands to a cautious starting range for quantised local language models. It also applies a little extra caution to longer-context, agent and vision workloads.
FREE · PRIVATE · BROWSER-ONLY
Enter two hardware numbers for a conservative first-pass estimate. Everything is calculated in this page. Nothing is uploaded and this is not a benchmark or compatibility guarantee.
Integrated/shared GPU memory is not the same as dedicated VRAM. If you have no dedicated GPU, enter 0.
Use the form to see a conservative hardware-fit range.
It maps broad RAM and VRAM bands to a cautious starting range for quantised local language models. It also applies a little extra caution to longer-context, agent and vision workloads.
Exact model architecture, quantisation, context length, runtime support, GPU backend, memory bandwidth, CPU speed and model quality all change the real result. A model may load and still be too slow for your workflow.
LLMRadar is built around hardware-aware model recommendations and local runtime evidence. HardwareRadar focuses on the Windows hardware and telemetry layer.