FlashDiff is a diffusion serving system that improves inference efficiency through adaptive regional execution and scheduling that reduces end-to-end serving latency by 30-97% and improves throughput by 1.2-2.2x.
Abstract
Diffusion models have become the central backbone for modern image, video, and audio generation, but their efficient service remains a challenge. Unlike autoregressive decoding, diffusion inference repeatedly updates high-dimensional spatial or temporal latents over many denoising steps. This all-region execution pattern makes generation latency high and limits serving throughput. Existing multi-GPU parallelization methods can reduce per-step computation, but often introduce substantial activation exchange overhead, causing communication to offset or even outweigh the benefits of parallel execution. This paper presents FlashDiff, a diffusion serving system that improves inference efficiency through adaptive regional execution and scheduling. FlashDiff is based on the observation that diffusion refinement is not uniform across latent regions or denoising steps: different regions often stabilize at different rates, while neighboring steps exhibit strong temporal correlation. FlashDiff leverages these properties to selectively execute only regions that require further refinement and to reallocate the resulting compute slack across concurrent serving requests. FlashDiff consists of three mechanisms. First, it decomposes the latent representation into coherent execution regions using early-stage attention signals, preserving semantic structure while exposing fine-grained parallelism. Second, it uses a lightweight runtime controller to estimate region activity and bypass low-impact updates when further refinement is unlikely to affect output quality. Third, it applies an affinity-aware online scheduler that co-locates dependent regions, balances residual load across GPUs, and reuses reclaimed compute capacity to improve serving efficiency. Across real-world image, video, and audio workloads, FlashDiff reduces end-to-end serving latency by 30-97% and improves throughput by 1.2-2.2x.
Deploying Video Diffusion Models (VDMs) on edge devices is appealing for localized and privacy-preserving generation, but their iterative Transformer-based denoising remains too slow for practical local inference. Cross-Timestep Caching (CTC) has emerged as a promising direction for reducing redundant computation, reusing activations across adjacent denoising steps rather than modifying model weights, while largely preserving generation fidelity. However, on memory-constrained edge GPUs, CTC requires a massive cache footprint that quickly exceeds on-device VRAM and forces the cache into host memory. More fundamentally, cache operators remain tightly interleaved and chain-dependent with native compute operators, so naive near-memory offloading still incurs repeated PCIe exchanges for residual and fusion computations, turning cache reuse into a communication- and serialization-bound execution flow. We therefore propose CODA, an algorithm-hardware co-designed architecture centered on Compute-Cache Operator Disaggregation. CODA separates dense compute paths and memory-bound cache paths across the xPU and a lightweight DIMM-side near-memory engine, reorganizes fragmented cache activity into hardware-friendly coalesced segments, and exploits Classifier-Free Guidance (CFG) branch independence to overlap xPU compute with cache-side execution. Experiments show that CODA achieves up to 1.80x end-to-end speedup and 1.74x higher energy efficiency, while preserving competitive generation quality compared with a state-of-the-art caching algorithm.
OmniCache is proposed, a unified hierarchical caching framework that performs multidimensional feature reuse through Token Cache, Frame Cache, Block Cache, and Layered Cache that reuses spatial features in temporal layers and temporal features in spatial layers, while Layered Cache captures cross-step redundancy at the model-layer level.
Zhaoyuan He, Muhammad Muaz, Lili Qiu· arXiv.org· 0 citations
Video diffusion transformers (vDiTs) generate high quality video but introduce extremely high compute cost due to the long diffusion timesteps and self attention computation. As diffusion timesteps are reduced, the computation cost of self attention becomes the dominant bottleneck. Existing acceleration approaches largely inherit sparse attention techniques from large language models, which fail to consider the unique spatiotemporal correlation of video data. This paper presents Kaleido, an algorithm hardware codesign that accelerates all operations in vDiTs by exploiting channel-wise spatiotemporal correlations in latent space. Based on this insight, we propose a lightweight channelwise reuse algorithm that skips redundant computations by reusing partial results while preserving higher generative quality than prior methods (>17 dB). To efficiently support this algorithm, we design a systolic array like accelerator with reconfigurable processing elements and a lightweight data dispatcher to mitigate irregular sparsity and data access patterns introduced by our reuse algorithm. Evaluations across three mainstream vDiT models show that Kaleido achieves up to 5.9x speedup and 16.0x energy savings over state of the art accelerators.
Wenxuan Miao, Haosong Liu, Weiming Hu et al.· arXiv.org· 1 citation
The exceptional generative capabilities of modern diffusion models are fundamentally bottlenecked by the quadratic computational complexity of their attention mechanisms. While recent feature caching strategies attempt to accelerate inference by skipping layers at static intervals, they fail to account for the non-linear evolution of latent features, inevitably causing severe structural distortions and temporal flickering. To address this, we propose AST-ToMe (Adaptive Step-Aware Thresholding), a novel dynamic gating mechanism that utilizes a runtime L2 norm feature drift metric to adaptively determine whether to compute or reuse attention states. Furthermore, we extend AST-ToMe to video stream generation through a cross-frame state inheritance design. Experimental results demonstrate that for single-image synthesis, AST-ToMe achieves a 15.3% reduction in inference latency with near-lossless perceptual quality (LPIPS: 0.0023). In continuous video generation, our method not only accelerates inference but also serves as a robust temporal anchor. By effectively suppressing random stochastic variations, AST-ToMe successfully reduces Temporal Jitter from 0.2059 to 0.1865, paving a highly efficient path for stable, flicker-free video generation.
Shu-Zhi Zheng· International Conference on...· 0 citations
LinCa decomposes cached features into sub-components with distinct continuity properties via a lightweight invertible network and applies differentiated prediction orders matched to each component, forming a unified Decompose-Predict-Reconstruct pipeline.
Jinshan Liu, Haoran Qin, Xiaobing Tu et al.· 0 citations
This work proposes Ms.Forcing, an efficient streaming video generation paradigm that adapts spatial granularity to each state's noise level and introduces Homogeneous-Noise-Level DMD, which assembles each fake video from clean predictions sharing the same source noise level, thereby reducing the mismatch between DMD training sequences and inference-time rollouts.
Zekun Li, Xiaoyan Cong, Hongyu Li et al.· arXiv.org· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
