Rhumb and local-hardware calculators
Hardware calculators tell you which open models fit your GPU or Mac: llmfit from the command line, CanIRun.ai in the browser. Others, such as tps.bunai.cc and ApX's calculator, estimate speed or size multi-GPU set-ups. Rhumb answers the next question in the same request: the best model for the task that fits, on the runtime you use, or a cloud model when nothing local does (see Local models).
| Aspect | Local-hardware calculators | Rhumb |
|---|---|---|
| What you ask | Does this model fit my hardware, and how fast will it run? | Which model fits this task within max_memory_gb, max_vram_gb and max_disk_gb? |
| Runtimes | A model file at a quantization. | Ollama, LM Studio, llama.cpp, MLX and WebLLM, with the id each runtime pulls. |
| Cloud | Not covered. | Local runtimes and your cloud providers in one ranking, so a laptop app can fall back to the cloud. |
| Quality | Not their job. | The same quality scores as cloud models. |
| Speed | Some estimate tokens per second. | Not given. |
When local-hardware calculators are the better fit
- You want a tokens-per-second estimate or to size a multi-GPU set-up, which speed calculators do.
- You've already picked the model and only need to know whether it fits.
What this page says about local-hardware calculators was checked on 2026-11-07 against their own pages. Products change; if something here is out of date, tell us.