Machine Learning on Agate
On March 27, 2022, MSI put its newest supercomputer, Agate, into full production and made it available to all MSI users. Agate is a heterogeneous high-performance Linux cluster with AMD processors, NVIDIA GPUs, and a high-speed communication network. All nodes have local solid-state scratch storage and AMD 7763 processors with 64-128 cores per node.
Before Agate was fully available, some of MSI’s biggest users were able to access the machine early to test its capabilities. One of these groups was that of MSI PI Ju Sun (professor, Computer Science and Engineering).
Professor Sun’s group develops modern machine-learning, especially deep-learning, methods to build robust image recognition systems, solve difficult scientific and engineering problems that defy traditional methods, and complement, inform, and augment doctors in digitalized healthcare. They also build novel theoretical and computational foundations for these methods. In short, this group works on the foundations and applications of modern artificial intelligence (AI). No single component of their research is free from GPU computing, and this group has been among the major users of MSI's GPU computing nodes.
On Agate, the better integrated GPU memory per card and large number of GPU nodes boost computing productivity by 10x. This improved turnaround means the researchers can develop AI methods at a fraction of the previous time needed. Also, GPU memory is always a bottleneck for experimenting with bold, exciting ideas. The much-expanded memory per computing node allows them to rethink and even shake the foundations of popular AI methods that they believe are suboptimal.
MSI users are welcome to begin using Agate to advance their research.
posted on May 13, 2022