Cornelis Networks Talks High Speed Fabrics for Heterogeneous HPC-AI

Print Friendly, PDF & Email

We caught up with Phil Murphy, CEO of fabrics technology company Cornelis Networks, which has one of the most interesting vendor histories in the HPC community.

Extending back to the 1990s and carrying forward extensive interconnect R&D by both Intel and Cray, Cornelis’s OmniPath is a fabric uniquely well-suited to the increasingly heterogeneous world of HPC-AI computing, Murphy said.

“The problem that we’re trying to solve in HPC and have been for a long time,” he said, “…has been not how do we get more applications on a single server, but rather, how do we get many servers working on the same problem at the same time. The fundamental problem that we’re trying to solve is to improve parallel computing, which has deep roots in modeling and simulation in the HPC world. But more recently, we’re seeing it crop up in other really important markets, for example, artificial intelligence based on machine learning.”

This, he said, involves more architectures, more programming models, more data – more of everything, particularly more complexity. In his comments, Murphy delivers an update on Cornelis’s progress in developing a fabric delivering high performance for this environment.