v4.6.0
“Can I Run This Model?” — an LLM VRAM calculator
July 9, 2026
Can I Run This Model? — will your GPU run that LLM
Local LLMs are everywhere, and the first question is always the same: will it fit on my card? This answers it with real memory math — and, like everything here, it runs entirely in your browser.
- Can I Run This Model? (
/tools/can-i-run-this-model) — pick a model (Llama, Mistral, Qwen, Gemma, Phi, Mixtral, DeepSeek and more, or type a custom parameter count) and your hardware (NVIDIA, Apple Silicon, or custom VRAM + bandwidth), and it computes the real VRAM requirement: model weights (parameters × bits-per-weight), the KV cache for your context length (correctly accounting for Grouped-Query Attention), and compute overhead — then tells you whether it fits. - Every quantization at a glance. A table shows FP16 down through Q8, Q6_K, Q5_K_M, Q4_K_M, Q3, Q2 and IQ2 — each with its size, whether it fits your card, a rough tokens-per-second estimate and a quality note — so you can pick the best trade-off. It highlights the recommended Q4_K_M and the best level that actually fits.
- Context, speed & Apple/CPU aware. A slider shows how context length eats VRAM (and you can quantize the KV cache to 8- or 4-bit to save memory); it reports the maximum context that fits and a memory-bandwidth-based speed estimate. Mixture-of-Experts models are sized on total params but run at active-param speed, and Apple unified memory and CPU + system-RAM are modelled too.
- 100% client-side. All the model specs, GPU specs and math are baked into the page — nothing you pick is uploaded, and it works offline.