Posted On: Mar 26, 2018
The AMIs now include an optimized build of TensorFlow 1.6 that is pre-configured with CUDA 9 and cuDNN 7 to take full advantage of mixed precision training on P3 GPU-powered instances. This improvement allows P3 instances to provide even greater training performance for machine learning applications.
In addition, the TensorFlow 1.6 build also now includes Intel's Advanced Vector Instructions (AVX, AVX-2, and AVX-512) to accelerate vector and floating-point operations on C5 instances. This TensorFlow build also takes advantage of Intel’s Math Kernel Library for Deep Neural Networks (Intel MKL-DNN) for CPU-acceleration of mathematical routines used in training deep neural networks, now fully configured in the AMIs.