Symphony of a Thousand - GPU Synthesis With Massively Parallel Oscillators

Explore GPU-based audio synthesis through this 38-minute conference talk from ADC 2024 that demonstrates building a real-time synthesizer using NVIDIA CUDA and massively parallel oscillators. Learn the design principles, implementation strategies, and practical tradeoffs required to create GPU-powered audio applications on consumer hardware, with a focus on wavetable synthesis algorithms that leverage GPU parallelism to outperform traditional CPU implementations. Discover how to integrate CUDA kernels with the JUCE framework to build a proof-of-concept synthesizer while addressing critical real-time challenges including CPU-GPU data transfer, latency management, buffering strategies, and GPU resource sharing with 3D graphics applications. Examine the complexities of real-time scheduling, persistent kernel design, and the practical considerations for deploying GPU audio applications on lower-end consumer hardware. Gain insights into identifying compelling use cases where GPU processing provides clear advantages over CPU-based audio synthesis, and understand the potential pathway toward standalone GPU hardware synthesizers for everyday musicians and audio developers.