Check what your PC can run
Enter your hardware below and CanIHost grades an entire model catalogue against your profile — fit level, memory requirement, estimated speed and the quantization that makes sense for your card.
Your system
Not sure about a value? A close estimate still produces a useful compatibility analysis.
Recommended for your PC
Prefer reading first? Browse the guides to understand what VRAM, quantization and fit levels mean before you check.
What the inputs mean
GPU and VRAM
VRAM is the memory on your graphics card, and it is usually the deciding resource: model weights live there during inference. If you are unsure of your card's VRAM, search the model name plus "VRAM" — 6 GB and 8 GB are the most common consumer sizes. Our VRAM guide explains how much you need per model size.
RAM
System memory matters in two ways: it hosts models when you have no GPU, and it receives offloaded layers when a model slightly exceeds your VRAM. 16 GB is the practical minimum for comfortable use; 24–32 GB makes experimentation graceful.
CPU threads
Enter your logical threads (physical cores × 2 with hyperthreading/SMT). Threads have less impact on speed than people expect — inference is bandwidth-bound — but they matter for prompt processing and offload scenarios.
Backend
The acceleration stack your machine will use: CUDA for NVIDIA cards, ROCm for AMD on Linux, Metal for Apple Silicon, CPU otherwise. The backend shapes which runtimes and quantization formats are available — see CUDA vs ROCm.
Reading the results
Each recommendation includes a fit label describing how the model's memory relates to your hardware, an estimated speed, and the quantization that fits your profile. These are estimates from the llmfit engine — not measured benchmarks — and actual performance varies with runtime, drivers and context size.
New to the terms? Start with what quantization means or the overview of how fit levels are calculated.