gFlow: Distributed Real-Time Reverse Remote Rendering System Model

Yixiao Xu, Yubo Li, Wanzhao Xu, Yicheng Gu, Yun Wang, Jiangyuan Ma, Zhengwei Qi

Abstract: Remote rendering has become increasingly important in the multimedia domain, driven by the rise of graphics-intensive applications and the need for efficient resource utilization. Traditional remote rendering focuses on scenarios where local devices lack GPUs, relying on remote machines equipped with GPUs for rendering. However, in emerging scenarios like autonomous driving, where local devices have high-performance GPUs and remote machines serve primarily as storage and management hubs, a concept we refer to as reverse remote rendering, academic research in this area remains limited. In this paper, we introduce gFlow , a distributed real-time reverse remote rendering system model that optimizes local GPU utilization by leveraging the VirtIO framework for efficient data transmission and remote control. Experiments show that our system achieved 10 times the FPS performance of the Linux approach x11 forwarding while maintaining comparable rendering quality to local processing in real-time remote game play. Additionally, gFlow demonstrates excellent scalability and resource utilization, making it a robust solution for distributed real-time reverse remote rendering.