TimTheJedi Thank you for reference. It is understandable that Intel is trying to compete with NVidia for NN processing and releases its own technologies. However, I am afraid, this is not that simple. It is not enough to just release new chip and hope everyone will use it, you need to make sure leading AI toolkits know how to use it and have compatibility with it. So far only NVidia managed to become de-facto technology for NN processing, with its CUDA library. Pretty much every NN framework supports CUDA now. But almost none have compatibility with its Radeon counterpart, as well as this new Intel NPU. Only few have support for Apple NPU since it has been around for some time already in iPhones.
We are not FaceBook, we cannot afford developing our own NN toolchain, we just use one of freely available ones. And it only supports CUDA. Hence there is CUDA compatibility, and no support for latest stuff from Intel. Though it uses things like AVX2, which very old, but still helps to speedup NN computations.
Hope that makes sense.
