Intel is continuing to push its AI strategy, with the latest announcement of PyTorch 2.0, a Python-based compilation stack. PyTorch 2.0 is designed to deliver much faster inference performance on Intel CPUs.

PyTorch 2.0's extended use of oneDNN, an open-source deep neural network library, is aimed at accelerating graph functions by automatically identifying graph partitions to be accelerated. It fuses comput-intensive operations in inference operations and in training use cases.

Graphic from Intel.