The Heterogeneous Computing group has supervised students from multiple universities in Norway, including the University of Oslo, Narvik University College, and the Norwegian University of Science and Technology. Our ongoing aim is to continue to supervise students at different levels, and we have several master's projects available. We have had many master's students through the years, and many graduees have continued to work at SINTEF, either as researchers, or as doctoral students. As a master student associated to us you will be integrated in our research group, including informal discussions at the foosball table.
The projects we present here are suggestions, and show some of the width of our previous and ongoing master projects. If you have specific desires, please contact us for an informal discussion. Common to all the projects described is that they will begin with an introductory study of existing literature, and state of the art. The projects will continue with development, implementation and testing of (new) methods. All the projects require candidates with good experience with programming and/or numerical methods. The projects are further possible to tailor to the specific knowledge of the candidate. The projects will, in general, end up with results worthy of publication.
Depending on the requirements of the projects, the students will get access to the same hardware resources as our researchers, including, a 5-GPU computer, Playstation3, and Cell BE Blade Center. As we work in close cooperation with both industry and universities our activities often involve both state of the art technology and demanding customer requests. This enables us to provide project of both academic and industrial interest.
Virtualizing CUDA for CNvidias CUDA API is the preferred choice for many GPGPU computations. However, not all computers are CUDA-ready, and the correct configuration of drivers and compilers is a major headache. It would be beneficial to provide virtualized CUDA environments where a small pool of GPUs could be used by several guest operating systems with only a small performance penalty.
This project will include the development of a Xen driver that handles communication through the hypervisor and a userspace library that communicates with the driver.Contact: Johan Seland
Explicit schemes for evolutionary PDEs.Explicit schemes for evolutionary PDEs maps well to GPU hardware and high speedups have been demonstrated for applications to shallow water flows, gas dynamics, and astrophysics. The project can take several directions: (i) explore central difference schemes on triangular grids, (ii) investigate the use hierarchical parallelisation using multiple GPUs, (iii) use of higher-order (WENO) reconstructions for shallow water and/or gas dynamics. Contact: Knut-Andreas Lie
Flow in porous mediaSINTEF has for several years developed advanced computational methodologies for computing flow in porous media with particular emphasis on applications in reservoir simulation and subsurface storage of CO2. The project can take several directions: (i) use CUDA or OpenCL on programmable GPUs to accelerate streamline simulation: the solution of 1-D transport problems along streamlines and gravity lines and/or pvt and flash calculations; (ii) use of GPUs for linear algebra, in particular, algebraic multigrid and/or multiscale methods, (iii) build support for GPU accelleration into the MATLAB Reservoir Simulation Toolbox. Contact: Knut-Andreas Lie
3D Volume VisualizationVolume visualization is increasingly becoming more and more important. The use of computationally demanding algorithms, such as ray-tracing and ray-casting, has recently become feasible for realtime visualization.
Visualization of smooth surfacesApplications ranging from computer games and simulators to industrial design rely on efficient rendering of smooth surfaces e.g. subdivision and B-Splines. Visualizing spline surfaces correctly and in real time is a vital part of all CAD software. Functionality available in new graphics hardware can be exploited to efficiently visualize such surfaces, both in real time, and with good quality.
Algebraic surfaces are often difficult to understand from their mathematical formulation. Visualization of such surfaces is important to broaden our understanding of such surfaces and their singularities.
Contact: Tor Dokken
Interactive Ray-Tracing of Volumetric Data using the NVIDIA OptiX engine
More information to come.
Published December 2, 2008