Intel oneDNN 1.5 Released With Non-x86 CPU Support, Initial Xe Graphics Support

Intel's oneDNN 1.5 deep neural network library has been released that is part of their oneAPI initiative and formerly known as MKL-DNN and DNNL.

With oneDNN 1.5 they have continued working on a wide variety of performance optimizations. There are some broad optimizations in v1.5 but particularly for newer Intel CPUs there should be even faster performance.

Also significant with oneDNN 1.5 is initial support for Intel Xe Graphics, including both Gen12 Tiger Lake integrated graphics and the DG1 discrete graphics card. Expect more Xe Graphics optimizations to come to oneDNN in future releases.

Also interesting is that with oneDNN 1.5 there is now support for 64-bit Arm (AArch64) and other non-x86 processors. However, the code is particularly tested and highly tuned for Intel x86_64.

More details on Intel oneDNN 1.5 via GitHub. Via the Phoronix Test Suite and OpenBenchmarking.org, running benchmarks on oneDNN is as easy as phoronix-test-suite benchmark onednn or if more broadly interested in Intel's software offerings to run phoronix-test-suite benchmark oneapi.