CUDA Research Center

The rapid technological development of specialty processors for graphics cards (graphic processing units, GPU), which were originally developed for graphing calculator computer games and CAD programs, has been observed with great interest. It is also desirable to use these to solve computationally intensive problems in science and technology.

The interest is understandable when one considers that a modern high-end GPU has a processing power of around 4 teraFLOPS (4 trillion floating point operations) per second. This is comparable to the computing power of a super computer from the 90's.

JOANNEUM RESEARCH recognized this trend early on and, in 2007, had already begun to explore the potential of GPUs in a variety of applications and build up an extensive knowledge base.

Thus, some basic methods of image processing such as feature tracking in videos (KLT tracker), or the acceleration of calculations of characteristic image features by a factor of 5-10 with the help of a GPU, were developed. These GPU-accelerated computing methods are successfully used in various applications including the real-time quality analysis of videos, logo detection, 3D reconstruction and highly accurate odometry.

In DIGITAL, CUDA (Compute Unified Device Architecture) is used by NVIDIA because this currently represents the most advanced and powerful technology for GPU programming. Because of its numerous successful GPU-related activities, DIGITAL was the first research institute in Austria to be awarded the title of “CUDAResearch Center” by NVIDIA. As a CUDA Research Center, DIGITAL has early and immediate access to information about the latest generation of graphics cards and receives development samples.

Development partner  NVIDIA