Student Immersion Project

In early 2019, the Stanford High Performance Computing Center (HPCC) began its Student Immersion Project (SIP) to assist with the federal grant writing, retrieval, preparation, and installation of the Shepherd and Yellowstone systems. Combining these systems' most integral components provided the HPCC with a hybrid cluster capable of faster and more efficient parallel computing. This hybrid cluster includes a high speed InfiniBand network and thousands of additional next-generation nodes that enable improvements to transfer speed, compute availability, and compute power as compared to our previous, now legacy, Certainty system. Through the program, students were able to rebuild the HPCC's vital infrastructure from the ground up, while maintaining services for academic research groups and partners. In this time, research groups were able to notice incredible changes in the speed and utility of the hybrid cluster as its availability increased. Now that the hybrid cluster has entirely replaced Certainty, students will continue to administer over the systems and search for innovative ways to allow the system to make use of all of its compute power. We are proud to offer these improved compute cycles to our research community and are excited to see the new research that can be done on our clusters.

Building an HPC System with IBM POWER9

Caetano Melone

In early January, the Stanford High Performance Computing Center received a couple IBM POWER9 machines to experiment with. I had the pleasure of setting up and testing the installation process for two powerful computers that are used in some of the clusters at the forefront of the TOP500 list. My goal for this project was to evaluate the set up process for these systems as well as the performance that it would deliver.

After going through weeks of testing, I came to the conclusion that POWER9 systems aren't as accessible as other machines that have access to utilities such as OpenHPC, which allows for easy package installation. Despite the additional time investment required to set up a POWER9 cluster, the performance is unparalleled for applications in Artificial Intelligence and Machine Learning, making them incredibly valuable machines for academics. I have written a setup guide designed for beginners. I also presented at the 2020 Stanford HPC-AI Advisory Council Conference on this exact topic: