Real-Time Parallel Computing Using GPUs
Anne Elster (NTNU)
All slides are available at
http://www.idi.ntnu.no/~elster/talks/elster-gpu-ppam2009.pdf
The need for computational power is ever-growing, but as we hit the frequency and power walls, parallelism becomes necessary in order to satisfy this need. This is especially true for real-time applications such as medical imaging, as well as other applications that require that a lot of calculations be done in a short period of time. Driven by the gaming market, Graphical Processing Units (GPUs) have recently become serious computational devices capable of several Teraflops. This has led them to be a very attractive target for real-time and other HPC applications. In this invited talk, I will discuss some of the recent accomplishments of the HPC-lab at NTNU where I have a large group of students working on GPU computing. Sponsored by NVIDIA, we have worked on developing several HPC applications for the GPU in collaboration with my colleagues in Computer Science, Petroleum Engineering and Medical Technology as well as industrial collaborators, including Fast/Microsoft, StatoilHydro and Schlumberger. An outline of the challenges with current GPU computing and future directions will also be discussed.