A detailed analysis maps the KLD (Kullback-Leibler divergence) of KV cache quantization for Qwen3.6-35B-A3B and Gemma4-E2B models. Results show q8/q8 quantization is nearly lossless on both models, while q4/q4 performs well on Qwen but causes severe degradation on Gemma. Turbo quantization variants show mixed performance, with turbo3 and turbo2 enabling extreme cache compression at significant accuracy cost.