Evaluating CPU, GPU, and FPGA performance in the context of modal reverberation: a comparative analysis

The vibration of acoustic systems can be represented through modal decomposition, reducing the problem to a set of harmonic oscillators. This study investigates the real-time performance of CPUs, GPUs, and FPGAs in implementing such models, focusing on the synthesis of large-scale modal reverberation. By leveraging their respective architectures, these processors are assessed for their ability to manage the high computational demands of modal synthesis. GPUs and FPGAs, known for their parallel processing capabilities, are evaluated alongside recent multi-core CPUs, which increasingly approach similar performance levels in handling such tasks. Through a series of platform-specific optimisations, this paper examines the maximum achievable mode count, latency, and processing efficiency for each platform in various real-time scenarios. Results indicate that while GPUs offer superior scalability, FPGAs achieve unparalleled latency performance, making them suitable for specific low-latency applications. CPUs, conversely, demonstrate unexpectedly high performance in smaller-scale applications. This work provides insight into the practical application of each processor type within real-time digital signal processing and suggests pathways for future research in hardware-based audio DSP.