Can you run AI models locally? VRAM requirements by model

Every guide includes a quantization VRAM table and a live GPU fit list computed from our hardware database.

Each model page below answers one question — can you run that model locally — with a VRAM table covering Q4_K_M to FP16 quantization, the KV-cache cost of longer contexts, and which GPUs from our product database run the model comfortably, at minimum, or need a second card. Parameter counts and architecture details are verified against Hugging Face model cards, never guessed.

How much VRAM do popular LLMs need locally?

Which GPUs run FLUX, SDXL, and Stable Diffusion 3.5 locally?

How much VRAM does local video generation need?

What GPU do Whisper and XTTS-v2 need locally?

How are these VRAM numbers calculated?

Weights are computed as parameters × bits-per-weight (Q4_K_M ≈ 4.8, Q5_K_M ≈ 5.7, Q6_K ≈ 6.6, Q8_0 ≈ 8.5, FP16 = 16), plus 10% loading overhead, plus a KV cache derived from each model's verified layer count, KV heads, and head dimension. The full method is documented on the methodology page, and you can run your own numbers with the VRAM calculator.