Dear Computing Community,
This week, we are lucky to have Inanc Senocak (ME, Univ. of Pittsburg), a former BSU faculty, come speak to us about his work in developing massively parallel GPU-based numerical simulations for fluid dynamics. His talk promises to be an interesting mix of high-performance computing, data compression via adaptivity, and computational fluid dynamics.
Speaker: Inanc Senocak (Mechanical Engineering and Materials Science, University of Pittsburg)
Title: Extreme-scale computing for fluid dynamics simulations
Date: March 11 (this Thursday)
Time : 10:30AM (MST)
Place: Zoom
Register in advance for this meeting
After registering, you will receive a confirmation email containing information about joining the meeting.
Abstract
Programmable graphics processing units (GPUs) have transformed supercomputing over the last decade. Today, many of the top supercomputers support heterogeneous computing on central processing units and GPUs. Despite a tremendous growth in raw computing power, design and implementation of parallel simulation software that can exploit the full potential of a top supercomputer remains a formidable challenge. To this end, we have undertaken two software efforts toward realizing extreme scale computing for fluid dynamics simulations. In the first effort, we revisit octree generation for extreme-scale problems and introduce the binarized octree generation technique for Cartesian mesh refinement around immersed geometries with deep levels of adaptation. The essence of the method is a strict adherence to the bitwise representation of an octree. A unique feature of the binarized representation is that it makes it easy to identify the neighbors of an element, including off-branch neighbors, without explicitly storing the connectivity information. In the second effort, we develop a massively parallel direct solver, PittPack, for the solution of the elliptic Poisson’s equation for pressure, which is the most time-consuming section of an incompressible flow algorithm. We pursue an MPI-OpenACC implementation in PittPack for massively parallel computations. We introduce a chunked pencil decomposition with two different communication patterns to achieve good scaling on the now-decommissioned, Titan supercomputer. Our largest computation had 1.1 trillion finite-difference points and deployed 16,384 Nvidia Tesla K20X GPUs.
Bio
Inanc Senocak is a William Kepler Faculty Fellow and an associate professor in the mechanical engineering and materials science department at the University of Pittsburgh. He obtained his PhD degree in aerospace engineering from the University of Florida in 2002. He is a fellow of the ASME and an associate fellow of the AIAA.
You can keep up with events at the Boise State Computing Website
And if you miss a talk, you can catch up on the Boise State Computing YouTube channel
See you on Thursday,
Donna
Donna Calhoun (Mathematics)
Catherine Olschanowsky (Computer Science)
Liljana Babinkostova (Mathematics)
Vicken Hillis (Human-Environment Systems)