Intel Releases oneDNN 3.0 In Advance Of Sapphire Rapids

[phoronix]

Phoronix: Intel Releases oneDNN 3.0 In Advance Of Sapphire Rapids

Intel overnight released oneDNN 3.0 as the newest major release to this open-source project for assisting in building deep learning applications. This oneAPI software component can already be used by PyTorch, ONNX, MATLAB, and other prominent software while the v3.0 release prepares it for future Intel hardware...

https://www.phoronix.com/news/Intel-oneDNN-3.0

[kobblestown]

I hope it check for CPU features rather than CPU model so it doesn't actively shun AMD CPUs. It's not like it hasn't happened before..

[oleid]

I hope it check for CPU features rather than CPU model so it doesn't actively shun AMD CPUs. It's not like it hasn't happened before..

Tensorflow aleady uses oneAPI by default. It says that it detected $CPU_FEATURES on start-up and that looks correct. Yet, like you said: with intel one can never be sure that a non-intel CPU is artificially slowed down. Happened before.