Task Partitioning for Computer Graphics in Heterogeneous Computing Systems

Task Partitioning for Computer Graphics in Heterogeneous Computing Systems

Bogdanov M.K. (ITMO, St. Petersburg, Russia)
Suvorov A.M. (ITMO, St. Petersburg, Russia)
Ivashechkina M.E. (ITMO, St. Petersburg, Russia)
Sultanov R.R. (ITMO, St. Petersburg, Russia)
Lyakh G.S. (ITMO, St. Petersburg, Russia)
Bulaev A.P. (ITMO, St. Petersburg, Russia)

Abstract

This paper addresses the challenge of efficient hardware resource utilization in computer systems equipped with multiple graphics processing units for computer graphics rendering. We propose a hybrid pipeline execution approach based on functional partitioning of computational tasks across heterogeneous GPUs, diverging from conventional spatial or temporal load-balancing strategies. Algorithms were designed and evaluated for three rendering techniques: procedural cloud simulation, shadow mapping, and ambient occlusion. By leveraging the DirectX 12 API in Explicit Multi-Adapter mode, cross-GPU communication was achieved without reliance on specialized hardware interconnects (e.g., SLI/CrossFire bridges). Experimental results demonstrate performance gains of up to 28% for shadow mapping, 11% for ambient occlusion, and 59% for cloud rendering workloads, contingent upon system configuration. These findings validate the efficacy of functional partitioning in heterogeneous GPU environments and highlight its practical applicability for consumer-grade hardware, including gaming PCs and laptops boarded with discrete and integrated graphics subsystems.

Keywords

hybrid rendering; multi-GPU systems; DirectX 12; computer graphics.

Edition

Proceedings of the Institute for System Programming, vol. 38, issue 3, part 3, 2026, pp. 161-176

ISSN 2220-6426 (Online), ISSN 2079-8156 (Print).

DOI: 10.15514/ISPRAS-2026-38(3)-43

For citation

Bogdanov M.K., Suvorov A.M., Ivashechkina M.E., Sultanov R.R., Lyakh G.S., Bulaev A.P. Task Partitioning for Computer Graphics in Heterogeneous Computing Systems. Proceedings of the Institute for System Programming, vol. 38, issue 3, part 3, 2026, pp. 161-176 DOI: 10.15514/ISPRAS-2026-38(3)-43.

Full text of the paper in pdf (in Russian) Back to the contents of the volume