Researchers at Harvard University's School of Engineering and Applied Sciences (SEAS) are focused on developing the hardware and software for exascale computers, while others are planning to apply exascale computing resources to diverse scientific fields once they become available.
An exascale computer would perform at least 1,018 operations per second, and a key challenge in realizing exascale systems is minimizing their electricity consumption. A SEAS research team investigating this issue found many design parameters must be optimized simultaneously rather than individually, while another researcher emphasizes improving the design of individual transistors and the materials from which they are manufactured.
Exascale systems likely will have to make do with less memory per processing core, unless new memory devices can be created. SEAS dean Cherry Murray expects heterogeneous computer architectures to dominate scientific computing in the coming years, using specialized subsystems optimized for different classes of algorithms. Also under consideration are systems specialized for one specific operation. Indeed, Institute for Applied Computational Science director Hanspeter Pfister sees a basic rethink of programming models as essential to an exascale transition. "We're beyond the human capacity for allocating and optimizing resources," he says. Pfister suggests shifting some concurrent computation onto hardware, while creating a new level of abstraction to spare coders from micromanaging parallel processes.
From Harvard Topics
View Full Article
Abstracts Copyright © 2014 Information Inc., Bethesda, Maryland, USA