Max Context Calculator: How Many Tokens Fit on Your GPU?
Exact KV-cache math from published model architectures and real downloadable weights. Pick a GPU, a model, and a quant — the calculator solves for the largest context where weights + KV cache fit.
How it works. KV cache per token = 2 (K+V) × layers × KV heads × head dim × 2 bytes (fp16 KV), read from each model's verified architecture. Weights come from real GGUF/safetensors file sizes; where no artifact row exists the weight is an architecture estimate and is labeled as such.