As the trend towards green technologies and increasing power costs continues, Supermicro celebrated their collaboration with Preferred Networks (PFN) to build the world’s most energy efficient supercomputer.
The MN-3 supercomputer achieved the first place ranking in the Green500 semi-annual industry assessment.
“The Green500 international recognition confirms that Supermicro delivers resource-saving, superior design, and high-reliability to the market,” said Charles Liang, the CEO and President of Supermicro.
The innovator in high-efficiency server technology achieved 15% higher efficiency compared to the previous Green500 record held by RIKEN in 2018. The MN-3 reached a record of 21.11 Gigaflops of performance-per-watt, delivering a total performance of 1.62 Petaflops. This was important for engineers, as performance-per-watt determines the cost of power and the cooling requirements to keep the system running.
After an exhaustive selection process, PFN partnered with Supermicro to develop the customised server, addressing a wide range of applications that require ultra-fast communications.
“Supermicro was excited to work with PFN on this exceptional system supporting machine learning and deep learning applications and energy efficiency,” added Mr. Liang.
The PFN solution is based on the Supermicro GPU server that utilises Intel Xeon CPUs as well as MN-Core boards developed by PFN for the training phase in deep learning.
“We can deliver outstanding performance while using a fraction of the power that was previously required for such a large supercomputer,” said Yusuke Doi, the VP of Computing Infrastructure at PFN.
The MN-Core boards were created after PFN determined they needed faster and more optimised solutions since the existing accelerators were not keeping up with customer demand.
The supercomputer required out-of-the-box thinking to be able to fit two CPUs, four MN-Core boards, 6TB of DDR4 memory, multiple GPUs, and interconnects enabling ultra-fast communications between the GPUs.
Following the completion of the supercomputer, PFN and Supermicro achieved a cluster of 48 servers, four interconnect nodes and five 100GbE switches, with a total of 2,080 CPU cores and housed in a 7U high rack-mounted unit.
This efficient design will enable accelerated deep learning algorithms, empowering PFN to design new applications that address customers’ pressing requirements and Service Level Agreements at reduced operating costs.
If you have a story or opinion to share, get in touch at editor@w.media. And get the latest by signing up to the W.Media Newsletter!
