Intel Issues A Slew Of Open-Source Software Updates For oneAPI

Intel's open-source teams have been issuing a slew of new packages in recent days.

Among the notable Intel open-source packages crossing my radar in recent days include:

OpenVKL 0.9 - Various VDB improvements, build system updates, various fixes, a variety of internal API changes, and also providing reference binaries now for Linux / macOS / Windows. This release also fixes an issue I had with OpenVKL for the last time I tried using it as a benchmark, so there is now also an OpenVKL test profile and we will be using it in our future benchmark articles where relevant.

oneDNN 1.3 - Formerly known as DNNL and before that MKL-DNN, this deep learning library has a variety of performance optimizations including prepping for Intel "Cooper Lake" Xeons. There is also performance improvements as already outlined. Our test profile for it has also been updated in preparation for future CPUs.

Embree 3.9 - Intel's ray-tracing kernel repository has been updated with support for round linear curve geometry, performance improvements, and other changes. Yep, we are updating our test of it for future comparisons.

Intel Open Image Denoise 1.2 - This denoising library supported by Blender and other programs now has neural network training code, support for specifying user-trained models at runtime, a denoising speed boost by 7~38%, and other additions. Our test profile has already been updated for this release.

OSPray 2.1 - This ray-tracing rendering engine now has a number of new features and fixes.

Presumably with the number of oneAPI component updates over the past week, a new oneAPI "release" is probably imminent.

These updates paired with the new oneAPI Level Zero tests makes for some fun oneAPI benchmarking ahead with Intel's oneAPI continuing to show a lot of potential and industry interest. These Intel open-source projects continue showing off well and performing great across the x86_64 Linux spectrum.