To main content

GPU Compute

GPU Compute

The Heterogeneous Computing group has a long history of working with cutting edge GPU compute. From the early beginnings with register combiners, through OpenGL shaders, to becoming a certified CUDA Research Center.

SPH on GPU

In the SCORE project we were responsible for GPU acceleration of the SPH simulations. In addition to this, we were heavily involved in the general SCORE code design and code development. The end result was an efficient GPU implementation supporting many complex simulations. 

 

Dambreak simulation with SPH on GPU
Simulating impact of copper bar using SPH on GPU

K. O. Lye, C. Dyken, J. S. Seland, F. O. Bjørnson, J. O. Busklein, T. Coudert, M. Føre, P. Klebert, A. Lavrov, B. Lund, J. O. Olsen, C. Pákozdi, P. Skjetne, W. Yang. Effective memory layout and accesses for the SPH method on the GPU, in: Proceedings 8th SPHERIC Workshop held in Trondheim (Norway), 2013. [ResearchGate]

Simulation on GPU

Shallow water simulation is performed by solving a hyperbolic conservation law. These kind of simulations are perticularly suitable for GPU-based implementation, because they can be solved by using explisite schemes. We have therefore developed a highly optimized Cuda implementation.

Medical Ultrasound

Medical ultrasound machines use advanced signal and image processing to produce the best possible image quality. GPUs is the prefferred computational resource due to their flexibility and floating point performance. We help our industrial partner to make good design choises and optimize their algorithms.