We're kicking off an occasional series of posts about how scientists and developers around the world are using the CUDA parallel computing architecture to advance their industries.
CUDA ' which enables GPUs to handle data in addition to pixels ' dramatically increases computing performance, enabling processing to be accelerated up to several hundred times. It is increasingly being utilized in such areas as image and video processing, computational biology and chemistry, fluid-dynamics simulation, medical imaging, and more.
This week, Allan P. Engsig-Karup of the Technical University of Denmark (DTU), and GTC 2010 speaker, shared details about his research on the OceanWave3D wave simulation model. OceanWave3D will help engineers create better designs for structures that exist in the ocean environment, like oil platforms and offshore windmills.
NVIDIA: Allan, tell us about your work at DTU.
I am responsible for teaching and research related to scientific computing. I teach on the order of 200 BSc, MSc and PhD students every year.
My collaborative research is focused on GPUs for applications requiring efficient PDE (partial differential equation) solvers and optimization algorithms, as well as development of performance profiling tools.
A project I am currently involved in, with my colleague Associate Professor Harry Bingham, is the continued development of a tool referred to as OceanWave3D ' for simulation of nonlinear and dispersive free surface flow in marine settings.
Allan P. Engsig-Karup, DTU
NVIDIA: How can this research be used in the real world?
Allan: Coastal and ocean engineers need to estimate the flow kinematics and design loads on human-made structures in the ocean, such as ships, oil platforms, offshore windmills and energy devices.
Of particular interest in this project are the windmills, where predictions are required for the maximum expected lifetime load on the structure. With OceanWave3D, there are potential significant benefits ' in conjunction with existing tools ' for more accurate and efficient prediction of wave kinematics and wave loads.
NVIDIA: How did you become involved in GPU computing?
Some of my collaborators and colleagues from the U.S. (Andreas Kloeckner, Tim Warburton and Jan S. Hesthaven) were among the first to present generally-applicable scientific results demonstrating the astonishing achievements that can be made with GPUs when you combine the insight of algorithms and hardware. This is a powerful combination for enabling new discoveries.
NVIDIA: What kind of results have you achieved with CUDA?
Recently, working with one of my MSc students, we achieved impressive scalability results for the parallel GPU implementation of OceanWave3D. These results were achieved by careful redesign of algorithms and implementation on the hardware (using CUDA C).
NVIDIA: Tell us about Denmark's national GPU laboratory.
In 2008, with my research group, a national GPU laboratory (GPUlab) was established ' the first of its kind in Denmark. At the end of 2009, we succeeded in getting a large national grant and hired a few people for full-time research. In early 2010, we started to truly gain momentum in this exciting area. GPUlab is equipped with Tesla C1060s/2050s as well as CUDA-based GeForce cards.
NVIDIA: What other interesting areas are you working on?
We recently initiated work on developing a new numerical tool for multi-scale modeling of turbulence phenomena in plasma physics. This model will help bridge a gap between theoretical physics and practical experiments in the development of large fusion devices. This is an example of where parallel computing is essential when the problems are sufficiently large due to the physical scales involved.
Previous CUDA Spotlights: