In order to increase computing performance, current and future computer architechtures are parallel. However, modern multi-core CPUs (2-16 cores) uses up to 80% of its resources on non-computational tasks. At the same time, dedicated stream accelerators that contain hundreds of lightweight cores are available. Such stream accelerators are designed for high computational throughput, and are typically used by traditional cores to offload resource-intensive operations. Most applications contain a mixture of tasks, some are best suited for multi-cores and others for streaming accelerators, and will ultimately perform best on heterogeneous architectures.
Heterogeneous computing aims to combine the parallelism of traditional multi-core CPUs and accelerator cores to deliver unprecedented levels of performance for simulation and visualization.
Working on designing and implementing algorithms on modern GPU architectures and leveraging our deep understanding of the hardware we are able to solve problems in many different application areas.[Read more]
We employ advanced visualization techniques to make interactive visualization of scientific datasets, both 2D and 3D, which due to their sizes make it quite a challenge to achieve real time frame rates.[Read more]
By offloading work into the cloud we are able to integrate the power of a supercomputer into mobiles devices such as laptops, phones and tablets, though making efficient use of scalable GPU resources in the cloud is a challenging research task.[Read more]
Johan Seland, Research ManagerSINTEF ICT, Dept. Applied MathematicsP.O. Box 124 BlindernNO-0314 Oslo, NorwayPhone: +4797181614More