“The Pawsey Center employs a cooling mechanism that uses the shallow aquifer beneath the Center to dissipate the accumulated heat generated by Setonix,” Stickells said.
“This eliminates the need to use cooling towers as the primary cooling mechanism, so there is no loss of water to evaporation, while there is enough solar power to run the required pumps.
“This means that Setonix’s cooling system has essentially no impact on the environment.”
The center’s cooling distribution unit takes heat away from the system through heat exchangers for the data center water, and the liquid cooling is delivered to the individual blades and components in the cabinet as part of a closed-loop system.
“The CDU requires inlet water temperatures of up to 32 degrees, which helps eliminate the need for chillers in many environments and further reduces energy usage,” Stickells said.
“Given the efficiency of liquid cooling compared to air, the power and cooling budget of a liquid-cooled supercomputer can be significantly lower than that of a similarly sized air-cooled setup.”
Setonix’s Green Credentials Earned It Ranked 4th on the Green500 list of supercomputersunveiled Tuesday at the SC22 conference in Texas.
It ranks among the supercomputers in France, the US, and Sweden, and can rightfully claim the crown as the most energy-efficient supercomputer in the Southern Hemisphere with a power consumption of 56.9 gigaflops/watts.
In addition to promoting green credentials, Setonix also Ranked 15th in the Supercomputer Top500 list.
Both rankings are based on the peak and maximum power of the Setonix GPU partition.
Stickells said the GPU portion of Setonix is about 80 percent more efficient than its CPU portion, making the entire infrastructure significantly more energy efficient than the supercomputer’s predecessor, the Magnus.
“One of the things we’re doing in the new Setonix system is actively evaluating the energy efficiency of the algorithms running on the system, reporting to researchers how much energy they’re using to complete their computational tasks,” he said.
“The energy efficiency of the hybrid system will give researchers the incentive they need to run algorithms on GPU infrastructure.
“So when researchers think about how they’re going to put their energy into using the system, they think about the advantages that they get by running on a GPU system, and then take advantage of that increased efficiency in their work.”
The Morning Newsletter is our guide to the most important and interesting stories, analysis and insights of the day. register here.