Open Image Denoise is one of Intel’s dozens of fantastic open-source components, including the numerous components that comprise their oneAPI software suite. For years, Open Image Denoise has been a fantastic, high-performance denoising library for ray-tracing applications. The program has long been CPU-based and very performant due to the use of current instruction set extensions. However, Open Image Denoise 2.0 was launched today, bringing GPU acceleration to Intel, AMD, and NVIDIA graphics processors.
GPU acceleration is the highlight of the major Open Image Denoise 2.0 release. This feature is analogous to other Intel oneAPI rendering components that have traditionally been based on CPUs but have since been extended to support GPU acceleration. This change was made so that modern Intel Xe Graphics / Data Center GPU Flex Series / Intel GPU Max hardware can leverage the capabilities of their hardware with this software.
SYCL support has been added to Open Image Denoise 2.0, and it can now be used on all Intel Xe DG2/Alchemist GPUs, ranging from the consumer desktop Arc Graphics all the way up to the Intel GPU Max series. In the same way, as many other components of Intel oneAPI do, in addition to supporting their own distributed GPUs (dGPUs), they have provided support for other graphics vendor products as well. In addition to supporting the SYCL back-end, OIDn 2.0 also supports NVIDIA CUDA and AMD HIP, making it possible for this denoising library to support hardware from several vendors.
In addition, the Open Image Denoise 2.0 version includes several new additions to the API, support for asynchronous execution, a physical device API for querying compatible devices on the system, and a number of other enhancements.
Through GitHub, you may obtain information regarding the Open Image Denoise 2.0 release as well as the ability to download the Apache 2.0 licensed source code or the Windows, macOS, or Linux binaries. OpenImageDenoise.org is the place to go if you are just now hearing about this fantastic open-source project for the first time. As time permits, I’ll be testing OIDn 2.0 on a number of different CPUs and GPUs to see how it works with Linux.