Case Study: Preferred Networks achieves #1 Green500 energy efficiency with Supermicro

Sophisticated Systems Supporting Deep Learning Projects at Preferred Networks Earn International Recognition

Preferred Networks (PFN), a Japanese technology company specializing in deep learning and robotics, faced a challenge where its existing infrastructure could not meet the performance and energy efficiency demands of its customer SLAs. PFN needed a new, highly efficient supercomputing solution to drastically accelerate deep learning training applications while addressing concerns over excessive power consumption and cooling costs. To achieve this, they required a partner with deep expertise in power-efficient system design.

Supermicro worked with PFN to create a custom solution based on their GPU server platform, designed to house PFN's custom MN-Core boards and Intel Xeon processors. This partnership resulted in the MN-3 supercomputer, which achieved the #1 ranking on the Green500 list as the world's most energy-efficient supercomputer. The system delivers 21.11 GFlops per watt and a total of 1.62 Petaflops of performance while using only 77 kW of power, significantly reducing operating costs and accelerating innovation for PFN.

Preferred Networks

Yusuke Doi

VP of Computing Infrastructure