A short guide on running TensorFlow (freshly released by Google in 2015) on AWS EC2 GPU instances. Rather than walking through the painful manual installation steps, the author shares a pre-built AMI (ami-cf5028a5) for the g2.2xlarge instance type in us-east-1, saving hours of setup. Known issues include inability to scale performance across multiple GPUs on g2.8xlarge. The post also briefly contextualizes TensorFlow as an incremental improvement over Theano, predicting that higher-level libraries like Keras will become the preferred interface for most deep learning work.
Sort: