GPU Audio is a company founded originally in Switzerland, that quickly evolved from a technology startup that received recognition in professional audio circles, particularly amongst the studio software world for music production, where programmers quickly realized the potential of developing plugins that could benefit from GPU acceleration, and the world of innovation happening in that space.
Through the world’s first fully capable solution for audio processing over GPU, and already with a permanent footing in Silicon Valley, GPU Audio expanded its approach to digital signal processing to encompass fast-growing applications in the automotive market. The company's unique tools offload DSP onto the parallel processing architecture of GPUs at 1ms latency round-trip, allowing for truly next-gen features, including remote computing and wireless/cloud processing.
Using the now-released GPU Audio SDK, developers can leverage all the benefits, including ultra-low latencies, multiple layers of processing, cross-platform support, and direct access to high-performance DSP. In addition, end-users will have the ability to run applications of multiple software vendors simultaneously on the same GPU.
The GPU Audio SDK has cross-platform support for Windows and macOS; with integration for NVIDIA and AMD GPUs as well as Apple Silicon chips. There's no need to write device-specific code for each platform, with each one running as low as 96 samples buffer or 96kHz sample rate on all target platforms (which results in a 1ms buffer).
A primary goal of the platform is to provide guarantees on backward compatibility. This enables developers, partners, and vendors to detach the update cycles of their products from the update cycles of the GPU Audio platform.
The SDK includes simple examples to get familiar with GPU Audio-specific APIs and create the first GPU-powered processor IIR and FIR processor - examples of IIR filtering and FIR/convolution. These are integrated into terminal/console tests that can be used to process files and measure performance. A Neural Amp Modeler with GPU acceleration of the real-time inferencing is another example, including everything to build a VST3 on Windows, with VST3 and AUv2 on macOS.
Use cases for the SDK are unlimited, but could include dynamic spatial reverbs, room simulation, acoustic room correction, multichannel sound synthesis, real-time inferencing of complex models and machine learning-enabled workflows (such as harmonizing multiple instruments) and more.
The first example of a professional audio use case included with the SDK is the mentioned NAM Plugin. The plugin itself is fairly CPU-intensive owing to its ML/NN basis, which is not a problem with GPU acceleration, allowing for improved performance and an overall number of simultaneous plugin instances running in real-time. Using the SDK, users will be able to build their own GPU-powered NAM Plugin and run it in any DAW that supports VST3 or AUv2. More use cases and example projects will be included in the coming months.
Programming your GPU
The GPU Audio SDK includes four components: GPU Audio Component (the audio processing engine), Processor API (used to write processors that run on the GPU), Engine API (which is used to load multiple processors, define processing graphs, and initiate processing on the GPU), and the DSP Components Library (GPU Primitives).
The SDK uses C++ on the host side and the device side, and all supported GPU vendors have a C++ dialect for their code. Software can be programmed using a C++ dialect that is the common subset of C++, CUDA, Metal, and potentially OpenCL.
The platform itself encapsulates the differences between a variety of GPU dialects through templates and a context object. The main difference to modern C++ is the additional keywords needed to decorate device memory spaces and GPU functions.
"As the demand for higher-fidelity, multichannel processing, and experiences grows, the use of GPUs for audio is a logical progression - and this SDK represents the first steps towards democratizing access to that previously untapped power," states Alexander Talashov, co-founder and CEO of GPU Audio.
Download the GPU Audio SDK here: https://gpu.audio/sdk
www.gpu.audio
