Diffusion Transformers (DiTs) are increasingly adopted in scientific computing, yet growing model sizes and resolutions make distributed multi-GPU inference essential. Ulysses sequence parallelism scales DiT inference but introduces frequent all-to-all collectives that dominate latency. Overlapping these with computation is difficult due to tight data dependencies, large message volumes, and asymmetric interconnect bandwidths. We introduce CoCoDiff, a distributed DiT inference engine exploiting two observations: (1) V requires only linear projection while Q/K need additional normalization and RoPE, creating opportunities to overlap V's communication with Q/K computation; (2) adjacent denoising steps produce similar tensors, yielding temporal redundancy. CoCoDiff introduces three mechanisms: Tile-Aware Parallel All-to-all (TAPA) decomposes collectives into topology-aligned phases; V-First scheduling hides V's communication behind Q/K computation; and V-Major selective communication transmits only active projections on slow interconnects. On the Aurora supercomputer with four DiT models across 1-8 nodes (up to 96 Intel GPU tiles), CoCoDiff achieves an average speedup of 3.6x, peaking at 8.4x.
@article{arxiv.2604.14561,
title = {CoCoDiff: Optimizing Collective Communications for Distributed Diffusion Transformer Inference Under Ulysses Sequence Parallelism},
author = {Bin Ma and Xingjian Ding and Tekin Bicer and Pengfei Su and Dong Li},
journal= {arXiv preprint arXiv:2604.14561},
year = {2026}
}