Skip navigation
Submerged Supercomputer Named World's Most Efficient System in Green 500
The Tsubame KFC system at Tokyo Institute of Technology immerses its servers and GPUs in cooling fluid. The system topped this year’s Green 500 list of the most efficient supercomputers. (Photo: NVIDIA)

Submerged Supercomputer Named World's Most Efficient System in Green 500

The Tsubame-KFC system at the Tokyo Institute of Technology turned in a record performance of 4.5 gigaflops per watt, placing atop the Green 500 by using a combination of CPUs and graphics processing units (GPUs) immersed in GRC's cooling fluid.

The Tsubame KFC system at Tokyo Institute of Technology immerses its servers and GPUs in cooling fluid. The system topped this year's Green 500 list of the most efficient supercomputers. (Photo: NVIDIA)

The Tsubame KFC system at Tokyo Institute of Technology immerses its servers and GPUs in cooling fluid. The system topped this year's Green 500 list of the most efficient supercomputers. (Photo: NVIDIA)

Green Revolution Cooling has touted immersion cooling as the most energy-efficient approach to high performance computing. Today it received a major affirmation when a Japanese system was named the world's most efficient supercomputer, placing atop the Green 500 by using a combination of CPUs and graphics processing units (GPUs) immersed in GRC's cooling fluid.

The Tsubame-KFC system at the Tokyo Institute of Technology turned in a record performance to place first , with an efficiency of 4.5 gigaflops per watt, marking the first time a system had ever achieved 4 gigaflops per watt or more. That was about 25 percent more efficient than the list’s number-two, Cambridge University’s Wilkes, at 3.6 gigaflops per watt. The Green 500 was created to spotlight the world's most energy-efficient supercomputers.

Tsubame-KFC uses two Intel Xeon E5 (Ivy Bridge) processors and NVIDIA K20 Kepler GPUs, which were immersed in a single CarnoJet cooling enclosure from Green Revolution.

Satoshi Matsuoka, a professor at Tokyo Tech, said the 4.5 gigaflops per watt was substantially improved from the to 3.1 gigaflops per watt for Tsubame 2.5 and less than 1.0 gigaflops per watt for Tsubame 2.0, which held world’s greenest title when it debuted in 2010. As a point of reference, Tianhe-2, the fastest supercomputer in the world according to the Top500 list, achieves an efficiency of 1.9 gigaflops per watt.

The November 2013 list marks the first time that all of the top 10 systems on the Green500 are heterogeneous systems - using computational building blocks that consist of CPUs, GPUs and co-processors. All incorporate Intel CPUs combined with NVIDIA GPUs. The Top500 newcomer Piz Daint at the Swiss National Supercomputing Center came in at number 4, with 1.7MW of total power, and 3.185 gigaflops per watt in energy efficiency. “A decrease in the average measured power coupled with an overall increase in performance is an encouraging step along the trail to exascale,” noted Wu Feng of the Green500.

Graph 500

A November 2013 Graph 500 list was also released - ranking the suitability of supercomputing systems for data intensive applications. The intent of the Graph 500 is to develop a compact application that has multiple analysis techniques (multiple kernels) accessing a single data structure representing a weighted, undirected graph. Backed by over 50 international HPC experts from academia, industry, and national laboratories, the Graph 500 will set large-scale benchmarks for big data applications.

IBM took the top 3 spots on the Graph 500, going to Lawrence Livermore National Laboratory’s Sequoia, Argonne National Laboratory’s Mira and Forschungszentrum Juelich’s (FZJ) JUQUEEN, which all use IBM Blue Gene/Q systems.  China's Milkyway-2, which ranked number 1 on the Top500, ranked number 6 on the Graph 500.

Hide comments

Comments

  • Allowed HTML tags: <em> <strong> <blockquote> <br> <p>

Plain text

  • No HTML tags allowed.
  • Web page addresses and e-mail addresses turn into links automatically.
  • Lines and paragraphs break automatically.
Publish