The open source Julia programming language has been admitted into the "Petaflop Club" by virtue of its Celeste application's peak performance topping one petaflop per second.
Developed by researchers at the University of California, Berkeley and their collaborators, Celeste processed the entire Sloan Digital Sky Survey astronomical image dataset via a new parallel computing technique.
The Celeste team loaded an accumulated 178 terabytes of image data to generate the most precise catalog of 188 million astronomical objects in 14.6 minutes with state-of-the-art point and uncertainty estimates. Celeste realized a peak performance of 1.54 petaflops using 1.3 million threads on 9,300 Knights Landing nodes of the National Energy Research Scientific Computing Center's Cori supercomputer, representing a 1,000-fold performance improvement in single-threaded execution.
Celeste's developers also are working to boost the precision of point and uncertainty estimates and to enhance the quality of native code for high-performance computing.
View Full Article
Abstracts Copyright © 2017 Information Inc., Bethesda, Maryland, USA