Splitting decode by attention type instead of by operator buys 31-56% more tokens per joule.

SQD argues disaggregated serving built around dense attention is the wrong shape for sparse, linear or sliding-window attention models, and splits decode by quadratic against subquadratic work. For sparse-attention models it separates top-k selection, which must index the full KV cache, from top-k attention plus FFN, which have static footprints. Measured on an adjusted 8xB200 proxy: 53% better tokens per joule on GLM 5.2, 31% on Nemotron 3 Ultra, 56% on Gemma 4 31B.