The notifications of ASC15 preliminary round has been just posted up. In the notifications, three brand new questions were shared with 152 teams from 135 universities.
ASC is one of the largest supercomputing challenges in the world, not only per its great popularity. It is also a challenge for teamwork and hands-on abilities. Even though many people are not so familiar with supercomputer, they still have great interests and curiosity on this frontier High-Tech. The contest questions, as a nice window, let us have a peek on this year’s session.
Speed to efficiency: HPL to HPCC
The first largest change in the ASC15 comes to the replacement of HPL with HPC Challenge (HPCC). HPL measures the floating point execution rate for solving a system of linear equations. While HPCC examines performance of HPC architectures using kernels with memory access patterns. The tests include four local (matrix-matrix multiply, STREAM, Random Access and FFT) and four global (HPL, parallel matrix transpose- PTRANS, Random Access and FFT) kernel benchmarks. Therefore, HPCC considers more features of the supercomputer, including flops, memory, and Ethernet. As proposed by Jack Dongarra, Piotr Luszczek and some experts from the U.S and Europe, HPCC is more comprehensive than HPL.
The proposition of HPCC also indicates the concepts changing from High Performance Computing to High Productivity Computing. We have already been aware that high performance alone can’t solve everything, one need to lean on the practical applications and increasing efficiency. This are the new requirements for supercomputer.
Molecule to Universe: NAMD + SKA
The other two questions are NAMD and Gridding, which study the microscopic and macroscopic world, respectively.
NAMD is a parallel molecular dynamics code designed for high-performance simulation of large biomolecular systems. It scales to hundreds of cores for typical simulations and beyond 500,000 cores for the largest simulations. NAMD has received 2002 Gordon Bell Award and 2012 Sidney Fernbach Award. NAMD is a very popular supercomputer code, and widely used in the field of life science, such the development of cancer medicine.
Another contest question, Gridding is a very important and time consuming computing procedure in SKA data processing. As the largest “big data” program in the world, the data produced by SKA every year could be up to 3000 PB, 30 times to that of Google.
To reconstruct the image of sky from the data collected by the radio telescope, scientists will have a set processes, one of which is that the irregular data produced have to be mapped onto the standardized 2-D mesh for the Fourier transformation.This process for mapping is called Gridding.
As introduced by Dr. Qing (Kitty) Ji, the executive director of ASC, the serial version of Gridding is still a big bottleneck of SKA data analysis. Those SKA scientists look forward to the parallelized and optimized Gridding code from the ASC15 challenge.
As illustrated by the three questions, there are two new features. First, adopting HPCC, ASC15 has paid more attention on the overall practical ability of the teams. Secondly, ASC15 is tougher than ASC14 as the complexity on the Xeon-phi’s porting and optimization of NAMD and Gridding.
“Try Different” is the slogan for ASC15. It refers to the difficulties and the significance of this year’s competition and has already been well embodied by the number of the registered team and the questions of the preliminary round.