GPU cluster projects

An acknowledgement of support from NCSA must appear in a publication of any material, whether copyrighted or not, based on or developed with NCSA resources. The suggested wording is:

"This work was partially supported by the National Center for Supercomputing Applications under [grant number] and utilized the [NCSA system name(s)]."

NCSA has worked with a number of researchers at the University of Illinois and other institutions to explore the use of GPU and FPGA technology to accelerate the performance of a number of science and engineering applications.

A General Relativistic Evolution Code on CUDA Architectures
Burkhard Zink, Center for Computation and Technology at Louisiana State University
Zink implemented a finite-difference code for solving Einstein's field equations on the system, comparing the resulting speed-ups to a serial code on a CPU for different parallelization and caching schemes.

CUDA-lite: Reducing GPU Programming Complexity
Sain-Zee Ueng, University of Illinois at Urbana-Champaign
NVIDIA's CUDA programming environment is an attempt to make programming GPUs more accessible. Difficulties remain, however, including dealing with the complex memory hierarchy. CUDA-lite is an automated tool to perform the transformations between memories.

Cactus, a software framework for high-performance computing
Erik Schnetter, Center for Computation and Technology at Louisiana State University
Cactus is an open source problem-solving environment designed for scientists and engineers. Its modular structure easily enables parallel computation across different architectures and collaborative code development between different groups.

GPU Acceleration of Molecular Modeling Applications
John Stone and Jim Phillips, University of Illinois at Urbana-Champaign
The parallel molecular dynamics program NAMD was ported to the NVIDIA CUDA GPU programming system and tested using NCSA's GPU cluster.

Investigating Application Analysis and Design Methodologies for Computational Accelerators
Volodymyr Kindratenko, NCSA
An investigation of the design methodologies for application implementation on FPGA and GPU-enabled high-performance clusters.

GPU acceleration of imaging algorithms for the Square Kilometer Array
Athol Kemball, NCSA/University of Illinois at Urbana-Champaign
The Square Kilometer Array is a next-generation radio telescope under current international cooperative design and development, and intended for phased construction in the decade 2010-2020.

GPU acceleration of Numerical Weather Prediction
John Michalakes, NCAR/University of Colorado, and Manish Vachharajani, University of Colorado at Boulder
Michalakes and Vachharajani are porting and optimizing the Weather Research and Forecast (WRF) model to the GPU cluster at NCSA to demonstrate that GPUs will provide dramatic improvement in speed, scaling, and power and cooling efficiency.

Understanding and analysis of video data
Thomas S. Huang, University of Illinois at Urbana-Champaign
The Image Formation and Processing group at the University's Beckman Institute is using the GPU cluster to explore real-time extraction of visual features in videos. The group hopes to build CUDA libraries that will enable many real-time computer vision applications that were thought to be too computationally expensive to be practical.

Kirchhoff Prestack Time Migration on GPU Cluster
Paulo Souza, Petrobras, Brazil
Kirchhoff prestack seismic migration is used in daily production at Petrobras. The goal of this experiment is to investigate the scalability on multi-GPU cluster.

The Effect of Spatial Localization on the Evolution of Biochemical Networks
Elijah Roberts, Leonardo Sepulvada, and Zan Luthey-Schulten, University of Illinois at Urbana-Champaign
Spatial localization of components inside the cell is an important factor in biochemical networks. Advances in imaging techniques have recently begun to reveal the precise 3D positioning of large biomolecules and GPUs now provide the computational power to perform simulations of large-scale 3D kinetic models.

High-Performance Simulation of Turbulent Flows on Graphics Processing Units
S.P. Vanka, University of Illinois at Urbana-Champaign
The goal of this work is to develop a robust, GPU-based computational fluid dynamics code capable of high-fidelity DNS/LES turbulence simulations. This code will solve the unsteady, 3D incompressible Navier-Stokes equations, and the algorithms will be written using CUDA (Compute Unified Device Architecture).