- About the Lab
- News and Media
- CSL Events
CSL Research Professor and Sanders-AMD Endowed Chair in ECE Wen-mei Hwu will participate in a three-year, $2.6 million grant from the Dept. of Energy as part of a team from Oak Ridge National Laboratory, the University of Oregon, and Georgia Tech to design a Next-Generation Software Infrastructure for Productive Heterogeneous Exascale Computing.
Dubbed “Vancouver,” the project will address multiple challenges on the road to an exascale high-performance computing architecture. The challenges include the persistent issues of realized performance, scalability and programming productivity.
Exascale computing will be faster and more energy efficient than petascale computing and may give the U.S. scientific community a global competitive advantage. “Vancouver” will involve porting applications such as simulations of new energy sources and more efficient engines for aircraft and vehicles from petascale to exascale platforms. The goal is to reduce the computation time from as much as six months to a year down to a few days.
A good part of the improvement will come from new computing devices, such as Graphics Processing Units (GPU). Hwu is a leading expert in designing and programming these devices. He recently led a joint UIUC-NVIDIA project for constructing and programming the EcoG GPU cluster computer that took the No. 3 spot on the November Green500 List. EcoG is one of the three most energy efficient supercomputers in the world today.
However, programming GPUs is currently difficult and time-consuming, which has been a major stumbling block. Hwu and Illinois faculty have been working to address the human side of the problem through courses and summer programs, which have drawn programmers from around the world. Hwu’s participation in Vancouver will focus on new programming tools that will greatly reduce the difficulty in programming GPUs and other types of parallel processors.
“If we are really successful, it will be a lot easier for scientists and engineers to port their applications to GPU-based systems, which can raise application performance by 20 times or more,” Hwu said. “Using these tools, scientists and engineers can instantly increase processing speed and reduce power consumption using the GPU hardware, without long programming delays.”