High Performance Computing

Sponsored by NVIDIA!

Session Flyer

Blue Waters – a supercomputer housed at the University of Illinois – was used to produce the most detailed EF-5 tornado simulations to date in 2017.

 

The High-Performance Computing session will focus on research dedicated to designing software to take advantage of today’s increasingly powerful computational hardware.  As we enter the era of exascale computing – an era in which processing power is expected to meet or surpass that of the human brain – developing code for science and engineering applications that fully and effectively exploits the massive amount of resources available is becoming increasingly vital, and research involving communication efficiency, performance modeling, GPU programming, and parallel algorithms sits firmly at the forefront of this effort.  This session aims to include talks that highlight the importance of numerical and computational tools that address these areas.

Summit – a supercomputer housed at Oak Ridge National Laboratory – performed the first exascale calculation while analyzing genomic information in 2018.

Keynote Speaker – Dr. Richard Vuduc, Computational Science and Engineering Department at Georgia Tech

Communication-avoiding Sparse Direct Solvers for Linear Systems and Graph Computations

 Abstract:

This talk describes several techniques to improve the strong scalability of a (right-looking, supernodal) sparse direct solver for distributed memory systems by reducing and hiding both internode and intranode communication. It also explains how to extend the same ideas to a graph problem, all-pairs shortest paths (APSP), exploiting the algebraic relationship between Gaussian elimination and APSP.The core idea is to reduce inter-node communication via a “communication-avoiding” 3D sparse LU factorization algorithm. The “3D” refers the use of a logical three-dimensional arrangement of MPI processes, and the method combines data redundancy with elimination tree parallelism. The 3D algorithm reduces asymptotic communication costs by a factor of O(\sqrt{\log n}) and latency costs by a factor of O(log n)for planar sparse matrices arising from finite element discretization of two-dimensional PDEs. For the non-planar case, it can reduce communication and latency costs by a constant factor. The methods also extend naturally to the case of heterogeneous CPU+GPU systems.

Bio
Richard (Rich) Vuduc is a Professor at the Georgia Institute of Technology (“Georgia Tech”). He works in the School of Computational Science and Engineering, a department devoted to the study of computer-based modeling, simulation, and data-driven analysis of natural and engineered systems. His research lab, The HPC Garage (@hpcgarage), is interested in high-performance computing, with an emphasis on algorithms, performance analysis, and performance engineering. He is a recipient of a DARPA Computer Science Study Group grant; an NSF CAREER award; a collaborative Gordon Bell Prize in 2010; Lockheed-Martin Aeronautics Company Dean’s Award for Teaching Excellence (2013); and Best Paper or Best Student Paper Awards at the SIAM Conference on Data Mining (SDM, 2012), the IEEE Parallel and Distributed Processing Symposium (IPDPS, 2015), and the ACM/IEEE Conference on Supercomputing (SC, 2018), among others. He has also served as his department’s Associate Chair and Director of its graduate programs.

External to Georgia Tech, he currently serves as Chair of the SIAM Activity Group on Supercomputing (2018-2020); co-chaired the Technical Papers Program of the “Supercomputing” (SC) Conference in 2016; and serves as an associate editor of the ACM Transactions on Parallel Computing (TOPC) and the International Journal of High-Performance Computing Applications, and previously for the IEEE Transactions on Parallel and Distributed Systems. He received his Ph.D. in Computer Science from the University of California, Berkeley, and was a postdoctoral scholar in the Center for Advanced Scientific Computing the Lawrence Livermore National Laboratory.

Student Speakers

 

Avatar Invited Speaker – Ramchandran Muthukumar, JHU
Randomized Sketching Algorithms for Low Memory Dynamic Optimization
Avatar Thiago Teixeira, UofI
A Language for Programming Optimization Spaces
Avatar Charbel Sakr, UofI
Finite Precision Deep Learning with Theoretical Guarantees
Avatar Jaemin Choi, UofI
Improving the Performance of Overdecomposed Applications on GPU-accelerated Systems
Avatar Ameya Patil, UofI
Boosted Spin Channel Networks for Energy-efficient Inference

Contact

For additional details, feel free to contact the session chair, Cory Mikida.