What makes a super computer become a super computer?
Only supercomputers have the power to answer questions such as these to gain powerful scientific insights
- Why does a hydrogen atom ionize when it collides with an electron?
- How to prove Big Bang theory?
- Where can the next earthquake strike?
Today, Supercomputers around the world are not only analyzing the past for scientific insights but are able to crunch BIG data sets to find solutions for a better world. The broad sets of applications that supercomputers execute have requirements that aren't consistent. Some applications require solving dense linear equations with regular memory access while others require sparse linear algebra computations with irregular memory access.
In the High Performance Computing (HPC) community the High Performance Linpack (HPL) benchmark is widely recognized and is used to optimize supercomputers for certain applications. Given the diversity of real world applications that supercomputers face it is critical to optimize them for benchmarks that are also diverse and closer to the real world applications. In that effort the High Performance Conjugate Gradients (HPCG) benchmark was created to represent sparse iterative solvers, which are often at the heart of the increasingly important class of irregular applications. A team of Intel researchers, led by Intel Lab's Parallel Computing Lab, has created an IA (Intel Architecture) friendly implementation of this benchmark, optimized for the latest Xeon and Xeon Phi-based supercomputing platforms.
This team has been actively collaborating over last few weeks with three leading supercomputing centers of the world: SuperMUC (Leibniz Supercomputing Centre in Garching, Munich), Stampede (Texas Advanced Computing Center) and Tianhe-2 (National Supercomputer Center in Guangzhou, China) to achieve the highest performance on their respective systems.
Today, a new ranking of world's supercomputer was announced based on this new HPCG benchmark (software.sandia.gov/hpcg/2014-06-hpcg-list.pdf), and the top spot on this maiden list goes to China's Tianhe-2 (TH2) supercomputer – one of the three beneficiaries of our research. TH2 has thus earned the unique distinction of leading both the traditional Top500 as well as the new HPCG ranking. SuperMUC, one of our other target machines, has earned the 6th spot in the HPCG ranking, while it is ranked 12th in the Top500 ranking. TACC's Stampede is listed at the 10th position on this list.
Scientists around the world compete for dedicated time and access to these supercomputers as they are trying to prove scientific theories. Even under such pressure our collaborators in these supercomputing centers have given us access and were available for us at unusual times.
Provided by Intel