Understanding 5G Performance on Heterogeneous Computing Architectures

Virtualized radio access network uses virtualization technology to enable multiple network modalities (e.g., different network services and QoS levels) to run on general-purpose computing devices instead of proprietary hardware. These devices include diverse types such as CPUs, GPUs, and DSPs. However, there is a lack of comprehensive performance comparisons across these heterogeneous computing architectures. This article aims to fill this gap by investigating the performance impact of parallel computing, I/O operations, and task orchestration on vRAN efficiency. Through empirical analysis, we explore the performance and architectural differences between CPU-based, GPU-based, and DSP-based solutions such as Intel FlexRAN, NVIDIA Aerial, and Marvell Octeon. Our study provides insights for optimizing 5G performance, and offers strategic recommendations for deployment tailored to different operational goals. We also outline future directions for research and development in vRAN architecture optimization.