"GRC is proud of its long history with TACC and we're delighted to have been able to collaborate once again with NVIDIA to help power the next generation of academic research", stated Peter Poulin, CEO of GRC. "We look forward to strengthening our partnerships with both NVIDIA and TACC to continue to provide support for this important academic endeavour."
First announced in 2018 and built earlier this year, the new supercomputer will enable the U.S. academic researchers to make important discoveries in all fields of science, from astrophysics to zoology, and further establishes the University of Texas at Austin's leadership in advanced computing. The GRC and NVIDIA subsystem is optimized for single-precision performance for TACC's multi-user environment.
"GRC and NVIDIA have developed a system that maximizes performance and efficiency", stated Dan Stanzione, Executive Director, TACC at the University of Texas-Austin. "GRC's immersion cooling will enable us to operate in a high-intensity computing environment without having to worry about performance, reliability or thermal issues."
GRC worked closely with TACC and NVIDIA to design and build custom servers that leverage the cooling capacity of GRC's ICEraQ solution, enabling TACC to maximize GPU density. NVIDIA's enterprise-class Tensor Core GPUs provide the computing power behind the systems accelerated computing nodes. GRC's ICEraQ system was chosen to support the GPU-intensive workloads due to its simplicity, virtually limitless cooling capacity, and energy efficiency.
Given the high density of the NVIDIA GPUs, motherboard and other components from Supermicro, and Mellanox InfiniBand cards for high-speed networking, using air-based cooling would have been complex and expensive. All available liquid-cooling options were considered and GRC's liquid-immersion cooling was chosen due to its ability to increase GPU density without having to plumb water lines to every individual heat source (GPUs and CPUs) within each server. By immersing complete servers in GRC's ElectroSafe dielectric coolant, GRC's ICEraQ delivers a consistent thermal environment for every server component, while dramatically reducing the required cooling energy.
"We're excited to collaborate with GRC and TACC to support the fifth fastest supercomputer in the world", stated Fred Allman, Director of Worldwide Supercomputing at NVIDIA. "NVIDIA and GRC have a long history of partnering, starting with the Tsubame KFC system which was no. 1 on the Green500 ranking in 2013 and 2014."
TACC first partnered with GRC in 2009 and validated the cost-effectiveness of its liquid-immersion cooling solution in both energy and operations. During the initial engagement, TACC found that it only took about 300 watts to cool 10 kilowatts of power, yielding a 1,03 PUE even in the middle of a Texas summer.