Implementing Real-Time Parallel DSP on GPUs

Explore the groundbreaking advancements in GPU-powered audio processing through this conference talk from the Audio Developer Conference 2022. Delve into the challenges and solutions of implementing real-time parallel DSP on GPUs, uncovering the potential for ultra-fast plugins, scalable power, and exponentially improved software performance. Learn about the three main challenges of GPU-based audio processing, including parallelism, multiple tracks and effects, and data transfer problems. Discover the core component of GPU audio processing - the Scheduler - and how it addresses these challenges. Examine practical examples of FIR and IIR algorithms and their parallelization potential, along with algorithmic and platform optimization techniques. Gain insights into the GPU Audio workflow, including components, APIs, and the DSP library. Conclude with a roadmap for future developments, use case considerations, and an invitation to a hands-on training lab for implementing gain, IIR, and FIR convolvers on GPUs.

Implementing Real-Time Parallel DSP on GPUs - Rumen Angelov & Andres Ezequiel Viso - ADC22