Chip manufacturers are constantly developing new and more powerful hardware architectures to stay on the exponential performance curve despite the breakdwon of Dennard scaling. Perhaps the most visible example of this is the transition to multicore technology in the general purpose-computing domain, but in some application domains - in particular graphics - many-core processors are already common. Other examples are SIMD vector extensions, which become wider with every generation as exemplified by Intel’s AVX-512, and the renewed interest in reconfigurable computing as demonstrated by the recent acquisition of Altera by Intel. Currently, however, there is little software that exploits these new technologies to their full potential. In this presentation, we will first briefly review the reasons for the fundamental shift to multicores. Thereafter, some of the research activities of the Embedded Systems Architecture research in these fields will be described. Examples of such activities are the development of ultra high performance video codecs, low-power parallel processing on GPUs, and the utilization of reconfigurable architectures for kernels that cannot be executed efficiently on software-programmable architectures. We will conclude with some challenges that need to be addressed in order to increase the employability of novel hardware architectures.
Sitzungszimmer des Dekanats der Fakultät für Informatik, Erzherzog-Johann-PLatz 1, 4. Stock