Japan's 'K computer' ranks No. 1 in four benchmarks at HPC Challenge Awards

The awards were announced on November 15 (US Pacific Standard Time) at SC11, the International Conference for High Performance Computing, Networking, Storage and Analysis taking place in Seattle, USA. The first-place rankings in the HPC Challenge Awards were received in the following four benchmarks used for evaluating the all-around performance of a supercomputer: 1) Global HPL; 2) Global RandomAccess; 3) EP STREAM (Triad) per system; and 4) Global FFT.

The HPC Challenge Awards demonstrate that, in addition to achieving successive top-place rankings on the June and November 2011 editions of the TOP500 list measuring LINPACK computational speed, the K computer is evaluated very highly in all-around performance as a general-purpose supercomputer. The K computer is currently under joint development by RIKEN and Fujitsu.

Background

RIKEN and Fujitsu have been working together to develop the K computer, with the aim of beginning shared use by November 2012, as a part of the High-Performance Computing Infrastructure (HPCI) initiative led by Japan's Ministry of Education, Culture, Sports, Science and Technology (MEXT).

The HPC Challenge benchmarks are benchmark programs designed to evaluate the overall performance of supercomputers in terms of processing performance in 28 tests derived from frequently-used computational patterns in the field of scientific computation. Among these, the four challenging benchmarks are: 1) Global HPL (operating speed in solving large-scale simultaneous linear equations); 2) Global RandomAccess (random memory access performance in parallel processing); 3) EP STREAM (Triad) per system (memory access speed under multiple loads); and 4) Global FFT (total performance of Fast Fourier Transform). The HPC Challenge Class 1 Awards are awarded to the top-ranked performance on each of these four benchmarks.

The University of Tsukuba contributed extensively to increasing the computational speed for the Global FFT benchmark. As a result, the performance results of the K computer were submitted to the Class 1 award category.

The HPC Challenge Class 1 Awards evaluate the performance of supercomputers from four different angles, and the K computer delivers world-class performance on all four benchmarks.

With the understanding that its use would be widely shared by researchers and engineers inside and outside RIKEN from the very start, the development of the K computer has proceeded with the aim of creating a supercomputer that combines superior computational performance with the versatility that enables it to run applications for a wide range of fields. The HPC Challenge results demonstrate the versatility of the K computer and the all-around high performance levels it delivers as a supercomputer.

Terms

1. K computer: The "K computer," which is being jointly developed by RIKEN and Fujitsu, is part of the High-Performance Computing Infrastructure (HPCI) initiative led by Japan's Ministry of Education, Culture, Sports, Science and Technology (MEXT). The K computer's availability for shared use is scheduled for 2012. The "K computer" is the nickname RIKEN has been using for the supercomputer of this project since July 2010. "K" comes from the Japanese Kanji character "Kei" which means ten peta or 10 to the 16th power. In its original sense, "Kei" expresses a large gateway, and it is hoped that the system will be a new gateway to computational science.

2. HPC Challenge Awards: The HPC Challenge Awards consist of the Class 1 benchmark performance competition and the Class 2 "Most Productivity" awards for the most "elegant" implementation of computationally intensive kernels. The Class 1 awards consist of the following four benchmarks, each of which evaluates the performance of key system components (CPU computational performance, memory access performance, network transmission performance).

• Global HPL: operating speed in solving large-scale simultaneous linear equations
• Global RandomAccess: random memory access performance in parallel processing
• EP STREAM (Triad) per system: memory access speed under multiple loads
• Global FFT: total performance of Fast Fourier Transform