[2603.01960] TiledAttention: a CUDA Tile SDPA Kernel for PyTorch

Computer Science > Machine Learning arXiv:2603.01960 (cs) [Submitted on 2 Mar 2026] Title:TiledAttention: a CUDA Tile SDPA Kernel for PyTorch Authors:Taimur Khan View a PDF of the paper titled TiledAttention: a CUDA Tile SDPA Kernel for PyTorch, by Taimur Khan View PDF HTML (experimental) Abstract:TiledAttention is a scaled dot-product attention (SDPA) forward operator for SDPA research on NVIDIA GPUs. Implemented in cuTile Python (TileIR) and exposed as a PyTorch-callable function, it is easier to modify than low-level CUDA templates while retaining realistic behavior via online softmax and tiled $K,V$ streaming. The approach is both performant and directly editable at the schedule level from Python (tile shapes, staging, shared-memory layout), enabling rapid, reproducible kernel research without template-heavy CUDA/CUTLASS rewrites. We benchmark TiledAttention on an NVIDIA DGX GB10 node with a reproducible harness and compare against PyTorch SDPA (auto-dispatch) and explicit unfused baselines across sequence length, head dimension, and precision (FP16/BF16). While production fused baselines remain stronger overall, TiledAttention delivers large speedups over standard eager attention paths and is available for direct use within PyTorch workflows, providing a practical balance between performance and customizability. Subjects: Machine Learning (cs.LG); Artificial Intelligence (cs.AI) Cite as: arXiv:2603.01960 [cs.LG]   (or arXiv:2603.01960v1 [cs.LG] for this version)   htt...