The National Science Foundation has awarded $10 million to the Pittsburgh Supercomputing Center (PSC), a joint research center of the University of Pittsburgh and Carnegie Mellon University, to build a new supercomputer. 

Called Bridges-2, the new computer will build on the most successful elements of its predecessor, Bridges. 

“Bridges-2 is a major leap forward for PSC in high-performance computing and data analytics infrastructure and research,” said Alan D. George, interim director of PSC and department chair of electrical and computer engineering at Pitt’s Swanson School of Engineering. “PSC is unique in combining the strengths of two world-class universities, CMU and Pitt, and a world-class medical center (UPMC). Bridges-2 will amplify these strengths to fuel many new discoveries.”

Like Bridges, Bridges-2 will be different from traditional supercomputers in two big ways.

First, instead of containing huge numbers of one type of computing hardware, it will contain different types of hardware for solving different types of problems across many disciplines. For computer scientists studying artificial intelligence, it will have state-of-the-art graphics processing unit nodes, which are particularly good at “deep learning” AI. For biologists who need to learn the DNA code of thousands of bacteria species in a soil sample without having to first isolate each species, it will have “large memory” nodes that can hold far larger amounts of DNA data than available in the nodes of traditional supercomputers. And for engineers studying coolant flow through next-generation nuclear power plants, Bridges-2 will support more traditional supercomputing with large numbers of smaller nodes.

Secondly, Bridges-2 will also be more user-friendly than a traditional supercomputer. The system will be tailored to “new community” researchers, such as those in the social sciences, digital humanities and medicine, whose fields had never needed computers before, let alone supercomputers. This will allow researchers to focus on their work, rather than having to learn how to program a supercomputer.