Close

Presentation

FuseDM: Accelerating Diffusion Model Sampling on Versal ACAP Architecture
DescriptionAs the diffusion model is a new trend in image generation applications, the fast sampling process of the diffusion model is a demanding development.
Meanwhile, Versal ACAP architecture provides both high multiply-accumulate throughput and efficient data transfer.

To accelerate the sampling process of the diffusion model on the architecture, we propose FuseDM, which performs fused operators on PL and AI Engines.
We also propose a performance model to analyze the workload of operators in the diffusion model.
In the experiments with one image sampling, our approach achieves $1.42\times$, $2.17\times$ faster than the Vitis-AI on VCK190 and Nvidia RTX 3080Ti, respectively.
Event Type
Work-in-Progress Poster
TimeWednesday, July 12th6:00pm - 7:00pm PDT
LocationLevel 2 Lobby
Topics
AI
Autonomous Systems
Cloud
Design
EDA
Embedded Systems
RISC-V
Security