@conference { , title = {There and Back Again: The Practicality of GPU Accelerated Digital Audio}, abstract = {General-Purpose GPU computing is becoming an increas-ingly viable option for acceleration, including in the audiodomain. Although it can improve performance, the intrin-sic nature of a device like the GPU involves data transfersand execution commands which requires time to complete.Therefore, there is an understandable caution concerningthe overhead involved with using the GPU for audio com-putation. This paper aims to clarify the limitations bypresenting a performance benchmarking suite. The bench-marks utilize OpenCL and CUDA across various tests tohighlight the considerations and limitations of processingaudio in the GPU environment. The benchmarking suitehas been used to gather a collection of results across vari-ous hardware. Salient results have been reviewed in order tohighlight the benefits and limitations of the GPU for digitalaudio. The results in this work show that the minimal GPUoverhead fits into the real-time audio requirements providedthe buffer size is selected carefully. The baseline overheadis shown to be roughly0.1ms, depending on the GPU. Thismeans buffer sizes 8 and above are completed within theallocated time frame. Results from more demanding tests,involving physical modelling synthesis, demonstrated a bal-ance was needed between meeting the sample rate and keep-ing within limits for latency and jitter. Buffer sizes from 1 to 16 failed to sustain the sample rate whilst buffer sizes 512 to 32768 exceeded either latency or jitter limits. Buffer sizesin between these ranges, such as 256, satisfied the samplerate, latency and jitter requirements chosen for this paper.}, conference = {NIME’20: Proceedings of the 20th New Interfaces for Musical Expression}, publicationstatus = {Unpublished}, url = {https://uwe-repository.worktribe.com/output/5951330}, keyword = {NIME, DMI, GPGPU, HPC}, author = {Renney, Harri and Gaster, Benedict R. and Mitchell, Thomas J.} }