The 2016 HPC Challenge Awards
Authors: Piotr Luszczek (University of Tennessee)
Abstract: The 2016 HPC Challenge Awards BOF is the 10th edition of an award ceremony that seeks high performance results in broad categories taken from the HPC Challenge benchmark as well as elegance and efficiency of parallel programming and execution environments. The performance results come from the HPCC public database of submitted results that are unveiled at the time of BOF. The competition for the most productive (elegant and efficient) code takes place during the BOF and is judged on the spot with winners revealed at the very end of the BOF. Judging and competing activities are interleaved to save time.
Long Description: The 2016 HPC Challenge Awards BOF is a competition with awards in two complementary classes: Class 1 and Class 2.
Class 1: Best Performance. The submitted results are selected on the basis of absolute performance numbers. The best run submitted to the HPC Challenge website is awarded with cash prize and a certificate. Since there are multiple tests in HPCC, the term "best" is subjective. It has been decided by the committee that the winners in four categories will be announced: HPL, Global-RandomAccess, EP-STREAM-Triad per system, and Global-FFT. Thus They represent important scientific kernels for the supercomputing community.
Class 2: Most Elegant. Awards an implementation of three or more of the HPC Challenge benchmarks with special emphasis placed on: HPL, Global-RandomAccess, STREAM-Triad and Global-FFT. This award would be weighted 50% on performance and 50% on code elegance/clarity/size. In addition, we allow presentation of additional codes (kernels and applications alike) that are critical to computational science.
Competition in Class 1 offers a rich view of contemporary supercomputers as they compete for supremacy not just in one category but in four. The four categories were chosen carefully to stress the variety of components of a modern high-end machines: the computational units (multicore processors and accelerators), memory hierarchy (caches and TLBs), as well as the interconnect fabric. The richness of information available from the submitted results allows for comprehensive analysis of the competing systems and it results in better understanding of both their strengths and weaknesses.
Class 2, on the other hand, offers a glimpse into high end programming technologies and the effectiveness of their implementation. The audience will have a chance to look at new programming and execution environments. The presenters are required to compare their solutions against the reference implementation and show advantages of their approaches in terms of programmer productivity and achieved performance. By using the existing HPCC kernels we allow for easy-to-follow introduction to new programming languages. And by allowing additional scientific kernels into the contest, we broadened the scope of Class 2 and made it more appealing to wider audience: both scientifically-minded and commercially-driven.
A brief introductory remarks precede the presentation of Class 2 contestants to give historical perspective and analyze the data from nearly a decade worth of HPCC and the HPCC Awards program. This includes performance trends over time as well as coming and going of parallel programming paradigms as featured by the statistics of submissions from prior years.
The judging is performed by a 10-person international committee of luminaries in the field of scientific and high performance computing. The judges themselves have been involved in numerous developments in their respective fields and lend years and decades worth of experience in hardware technologies and software environments as well as upcoming trends in supercomputing arena.
Birds of a Feather Index