Intel Optimised TensorFlow - AshokBhat/ml GitHub Wiki

About

• Optimizations: Using oneDNN is part of upstream TensorFlow.
• Enabled by default: No

Availablity

• Intel published TensorFlow binaries
• Docker and Python Wheels

Intel Docker Hub

• Docker Hub: https://hub.docker.com/r/intel/intel-optimized-tensorflow

Tag	Configuration
1.15.2	with oneDNN
2.3.0	With oneDNN
2.3.0-avx512-mkl	With MKL and no oneDNN

Google DL containers

• Google released Python3 DL containers for TensorFlow on CPU
• Format "tf-cpu.<framework version>"
• Starting version 1.14, optimized with Intel MKL DNN by default.
• https://cloud.google.com/ai-platform/deep-learning-containers/docs/choosing-container

Tag	Configuration
tf-cpu.1-15	TF 1.15
tf2-cpu.2-0	TF 2.0
tf2-cpu.2-3	TF 2.3