< Back to previous page
Project
A combined GP-GPU/FPGA desktop system for accelerating image processing applications (GUDI) (IWTTETR11BA)
Recently, two new technologies came into the market of desktop computer: the GP-GPU or general purpose graphical processing unit and FPGAs or Field Programmable Gate Arrays. GP-GPUs are powerful for massively strong, independent parallelism. FPGAs on the other hand are powerful for tasks with less regular parallelism, which also take into account dependencies. Realistic applications possess both forms of parallelism and this project will integrate both into a desktop computer. The right selections will be made in order to obtain optimal data transfer between GP-GPU , FPGA and processor.
The first objective in this project is the development of a generic development platform and tool chain based on hybrid desktop computers equipped initially with only GP-GPUs or FPGA computing boards. In the second phase, the ability to combine both technologies, is investigated. Existing basic primitives for medical imaging will be accelerated by combining GP-GPU/FPGA. Furthermore, a comparative study will be made for performance, scalability and development effort between GP-GPUs and/or FPGA solutions. Finally, further optimizations of algorithms will be carried out by splitting the algorithm into parts executing on the best performing platform. In this way, strengths of both technologies will be detected. Throughout the project, cases are used in the field of medical imaging in conjunction with the user comitee.
The first objective in this project is the development of a generic development platform and tool chain based on hybrid desktop computers equipped initially with only GP-GPUs or FPGA computing boards. In the second phase, the ability to combine both technologies, is investigated. Existing basic primitives for medical imaging will be accelerated by combining GP-GPU/FPGA. Furthermore, a comparative study will be made for performance, scalability and development effort between GP-GPUs and/or FPGA solutions. Finally, further optimizations of algorithms will be carried out by splitting the algorithm into parts executing on the best performing platform. In this way, strengths of both technologies will be detected. Throughout the project, cases are used in the field of medical imaging in conjunction with the user comitee.
Date:1 Jan 2011 → 31 Dec 2012
Keywords:industrial applied sciences, industrial sciences
Disciplines:Mathematical sciences and statistics, Civil and building engineering, Other engineering and technology
Project type:Collaboration project