October 2019 Summaries

This tutorial provides a step-by-step guide on setting up a Mellanox InfiniBand switch, specifically the SB7800 36-port EDR model. The subnet manager plays a crucial role in discovering and configuring devices on the InfiniBand fabric. To set up the subnet manager, users can either enable it via the command line or the web browser. For the command line approach, users need to perform an arp-scan to determine the switch's IP address, SSH into the switch, and then execute specific commands to enable the configuration terminal and the subnet manager. Alternatively, users can navigate to the switch's IP address in a web browser, log in with admin credentials, visit the IB SM Mgmt tab, click on Base SM, check the checkboxes for enabling the switch name and SM Enabled, and finally apply the changes.

The key points of the text cover building a simple CIFAR-10 image classifier using TensorFlow 2.0's Keras API, setting up a data pipeline with the dataset API, training the model with multiple GPUs and adding callbacks for monitoring progress and updating learning schedules. The tutorial demonstrates how to define a model using the Sequential API and functional API, set up a data pipeline, train the model, accelerate training speed with multiple GPUs, add callbacks for monitoring progress, updating learning schedules, and evaluate the model's performance.