Intel's oneDNN Continues Improving Support For Non-Intel Hardware

Earlier this year was the surprising move of Intel's oneDNN neural network library adding AArch64 support and that was then complemented by adding IBM POWER support to this neural network library that is part of their oneAPI collection. Now with the latest oneDNN 2.0 beta they have furthered the support and performance for non-Intel hardware.

Not only is there IBM POWER (PowerPC 64) support but IBM z (390x) is also now supported by this library formerly known as MKL-DNN and DNNL. This library is focused on providing the "building blocks" for constructing deep learning applications.

On the AArch64 front, they have gone beyond the basics and are now even supporting the Arm Performance Library (ArmPL) and Arm Compute Library (ArmCL) for offering optimized GEMM and convolution implementations for the 64-bit ARM architecture.

It's quite interesting to see Intel's open-source projects continuing to extend their support beyond just Intel (or x86_64) hardware. Once oneDNN 2.0 has been officially released, it will be fun to take it for some oneDNN benchmarking on these alternate architectures to see how they compete with Intel's optimized CPU support.

Wednesday's oneDNN 2.0 Beta 10 release also adds various Xe Graphics optimizations, more Intel CPU optimizations, new capabilities for CPUs and GPUs, dilated pooling support, and various other features.

More details on this newest oneDNN 2.0 pre-release via GitHub.