Georgia Tech’s latest parallel computing research is being presented at the SIAM Conference on Parallel Processing for Scientific Computing (SIAM PP18), at Waseda University in Tokyo this week, March 7-10. School of Computational Science and Engineering (CSE) Associate Professor Rich Vuduc is serving as the current chair of the SIAM Activity Group on Supercomputing, which is co-sponsoring the SIAM PP conference alongside the Japan Society for Industrial and Applied Mathematics this year.

A large number of CSE faculty members and Ph.D. students have research accepted for this year’s program and will be presenting during the week’s meetings. Of the 17 research presentations from Georgia Tech at the conference, 13 include presentations and work by CSE researchers.

Parallel computing allows for larger problems to be divided into smaller ones, which then can be solved simultaneously. This type of computation is inherent to high-performance computing (HPC), but is gaining more notice in other areas of computing research because they are increasingly able to take advantage of frequency scaling, a computer architecture technique used to conserve power and reduce heat generated by the chip.

"Parallel processing continues to have an increasingly big impact on physical simulation, data analytics, and artificial intelligence. The talks by Georgia Tech researchers reflect this with their particular focus on how to fundamentally advance the algorithms, data structures, and numerical methods underlying these domains,” said Vuduc.

Some of the themes emerging in this year’s work from Georgia Tech researchers demonstrate a range of parallel processing abilities and applications that stem from four main areas: data and HPC, novel systems, algorithms and libraries, and large-scale simulations.

Data and HPC

MS9, MS20, MS34, MS38, MS56, MS58, MS73

Novel Systems

MS30, MS38, MS41, MS97

Algorithms and Libraries

MS9, MS14, MS20, MS34, MS56, MS58, MS68, MS73, MS87, MS109, CP11

Large-scale Simulations

MS102, MS50, MS68

 

The following sessions include presentations by CSE faculty and Ph.D. students.

 

MS9: Tensor Decomposition for High Performance Data Analytics – Part I of III

HiCOO Hierarchical Storage of Sparse Tensors

Jiajia Li, Jimeng Sun, Rich Vuduc

MS20: Tensor Decomposition for High Performance Data Analytics – Part II of III

Organizer

Rich Vuduc

MS34: Architecture-Aware Graph Analytics – Part I of II

Scalable Graph Alignment on Modern Architectures

Ümit V. Çatalyürek, Bora Ucar, Abdurrahman Yasar

MS38: Deep Learning from HPC Perspective: Opportunities and Challenges – Part II of II

Faster, Smaller, and More Energy-Efficient Inference Using Codebook-based Quantization and FPGAs

Mikhail Isaev, Jeffrey Young, Rich Vuduc

MS41: Performance Engineering from the Node level to the Extreme Scale – Part II of II

Designing an Algorithm with a Tuning Knob that Controls its Power Consumption

Sara Karamati, Jeffrey Young, Rich Vuduc

MS56: Tensor Decomposition for High Performance Data Analytics – Part III of III

Organizer

Rich Vuduc

MS58: Scalable and Dynamic Graph Algorithms

Graph Analysis: New Algorithm Models, New Architectures

David A. Bader, Oded Green, Jason Riedy

MS68: Challenges in Parallel Adaptive Mesh Refinement – Part 1 of III

Structured and Unstructured Adaptivity in PETSc

Toby Isaac

MS73: Theory Meets Practice for High Performance Computing – Part II of II

Faster Parallel Tensor Compression using Randomization

Casey Battaglino

MS87: Innovative Methods for High Performance Iterative Solvers – Part I of II

ParILUT – A New Parallel Threshold ILU

Edmond Chow

MS97: Emerging Architectural Support for Scientific Kernels – Part I of II

Sparse Tensor Decomposition on EMU Platform

Srinivas Eswar, Jiajia Li, Richard Vuduc, Patrick Lavin, Jeffrey Young

MS102: Large-Scale Simulation in Geodynamics – Part I of II

Thermal Inversion in Suduction Zones

Toby Isaac

MS109: Techniques for Developing Massively-Parallel Linear Solvers – Part I of II

ACHILES: An Asynchronous Iterative Sparse Linear Solver

Edmond Chow

CP11: Algorithms

Chair

Ümit Çatalyürek

 

The following sessions include presentations by Georgia Tech faculty and Ph.D. students.

 

MS14: On Batched BLAS Standardization – Part II of II

Batched DGEMM Operations in Density Matrix Renormalization Group

Arghya Chatterjee

MS30: Resilience for Extreme Scale Computing – Part III of IV

Resilience with Asynchronous Many Task (AMT) Programming Models

Vivek Sarkar

MS50: Parallel Simulations in Life Sciences

Meshfree Simulations of Complex Flows Using General Finite Differences

Yaroslav Vasyliv and Alexander Alexeev