CUDA Center of Excellence

Georgia Tech has been designated by NVIDIA Corp., the world’s leading manufacturer of the graphics processing unit (GPU), as a CUDA Center of Excellence (CCOE).

What is CUDA? It is NVIDIA’s parallel computing architecture that enables dramatic increases in computing performance by harnessing the power of a GPU. With millions of CUDA-enabled GPUs sold to date, software developers, scientists and researchers are finding broad-ranging uses, including image and video processing, computational biology and chemistry, fluid dynamics simulation, CT image reconstruction, seismic analysis, ray tracing, and much more.

Georgia Tech is engaged in a number of research, development and educational activities that leverage GPU computing. These activities span the full gamut: applications, software development tools, system software and architectures.

Georgia Tech joins a select group of 10 other universities and research organizations in the United States and abroad, including Harvard University, Cambridge University and the Chinese Academy of Sciences, that are designated as a CUDA Center of Excellence. More than 350 universities worldwide teach the CUDA programming model within their curriculum.

As a CCOE, Georgia Tech will be engaged in a number of activities that promote research into GPU capabilities. Center leadership has a research vision that advances the state of applications of NVIDIA technology and fundamental computer science and engineering topics that may provide future enhancements to the use of this technology. Surprisingly, these activities cover virtually every dimension of scalable computing with heterogeneous computing with graphics processors: large scale computing facilities, education, algorithms and applications, architectures, libraries, system software, programming productivity, and performance.

One concrete example of Georgia Tech’s GPU research is the Keeneland Project. Housed at the Oak Ridge National Laboratory (ORNL), the National Science Foundation Track 2 Keeneland project aims to bring GPU computing resources to bear on NSF’s important computational science applications. Keeneland is a partnership between Georgia Tech, the National Institute for Computational Sciences, the University of Tennessee at Knoxville, ORNL, Hewlett Packard and NVIDIA, that will design, deploy and operate two heterogeneous computers. The initial delivery system, scheduled for summer 2010, will be an Intel-based Linux cluster with hundreds of NVIDIA Fermi graphics processors. A final delivery system, to be deployed in 2012, will use next-generation technologies at all levels and provide the open-science community with well over 1 PF/s of computational capability.

Georgia Tech also will coordinate GPU research by an impressive list of contributing partners, including:

  • Centers for Disease Control & Prevention (CDC), a major operating component of the Department of Health and Human Services, which provides expertise, information, and tools that people and communities need to protect their health. CDC is currently working toward implementation of several technologies for efficient and cost-effective supercomputing. This effort includes research and development on the use of GPU personal supercomputing platforms as well as GPU-enhanced HPC clusters.
  • Georgia Tech Research Institute (GTRI), the nonprofit applied research arm of GT with clients across industry and government. GTRI engages in a variety of GPU computing related research with a particular focus on usability infrastructure; it has been doing experiments, custom ports/implementations ranging from microkernels, compound functions, to full application implementations, since about 2003.
  • Oak Ridge National Laboratory (ORNL), a multidisciplinary, Department of Energy National Laboratory at the forefront of supercomputing and home to the world’s largest supercomputer.
  • Georgia State/Georgia Tech Center for Advanced Brain Imaging, which employs state-of-the-art methods, functional magnetic resonance imaging, brain stimulation, scalp electrical recording and sophisticated behavioral measures to understand brain function.
  • Accelereyes, a local Atlanta company that builds programming tools for parallel programming and visual computing on GPUs. AccelerEyes develops and markets Jacket, a software platform that is in use in dozens of countries by hundreds of engineers, scientists and analysts. AccelerEyes’ approach to application development and deployment empowers these domain experts to transform their serial applications, simply, into high performance NVIDIA GPU CUDA-enabled versions.

The existing research projects and education activities under way at these organizations demonstrate the continued interest and benefits in massively parallel computing and NVIDIA technology. By focusing these efforts under a CCOE, each participating organization can amplify its activities by cross-fertilizing ideas and skills internally, by sharing software and hardware facilities, leveraging training materials and efforts, and by streamlining interactions with closer, priority access to NVIDIA staff and capabilities.