NVIDIA CUDA Research Center
The Technische Universität München was awarded an NVIDIA CUDA Research Center in 2010, based on its ongoing activities in high performance computing using CUDA. The Technische Universität München is pushing towards GPU-based computing using CUDA for the purpose of large-scale visual computing, computational steering, and computational graphics. In a number of applications ranging from numerical simulation and image registration to data analysis and sample-based graphics, CUDA has been used to exploit the ever growing power of high performance multicore GPUs.
In one of the most recent projects at the Technische Universität München, CUDA has been employed to enable computational steering. Computational steering describes the interactive modification of objects and parameters during the run time of a simulation. It enables the user to modify the input parameters of the simulation and study the result on-the-fly. Thus, users in a computational steering environment obtain an intuitive understanding of the influence of the input parameters on the simulated processes.
To achieve this, the Technische Universität München has recently announced the world’s most efficient multigrid FE solver for linear elasticity using CUDA [CUDA-FE]. In close collaboration between the Computer Science Chairs for Computer Graphics and Visualization, High Performance Computing, Computer Architectures, and Augmented Reality at the Technische Universität München, this work is currently being extended towards the first fully immersive virtual windtunnel for real-time fluid simulations.
In another recent project, TUM researchers have demonstrated an efficient image compression library using CUDA. Even though image compression has traditionally not been seen as a GPU-friendly problem, the researchers could show an order-of-magnitude improvement in compression throughput by using a GPU compared to a highly-optimized JPEG2000 implementation on the CPU. The CUDA compression library has been used in an editing system for very high resolution terrain data sets, allowing the manipulation of both color and height.
A Real-Time Multigrid Finite Hexahedra Method for Elasticity Simulation using CUDA
C. Dick, J. Georgii, R. Westermann, Simulation Modelling Practice and Theory 19(2):801-816, 2011