Multicore designs are fast becoming a mainstream solution for High Performance Computing (HPC) systems. At the same time they are being supplemented by various hybrid architectures that can accelerate certain types of operations via customized hardware. These new systems present many new opportunities but also challenges to numerical software. New approaches are needed to effectively program today's four-core systems as well as upcoming systems with as many as 80 cores.
The changes in microprocessor architecture, initiated by the introduction of new multicore and hybrid designs, will inevitably work a revolution in the practices and programming models of the scientific software community. The new era in system design is being driven by the convergence of two difficult problems: physical barriers to further increase in processor clock rates, the widening of the gap between processor and memory performance. It is rapidly becoming clear that these changes will also render obsolete much of the software built on the old model, including fundamental building blocks of Computational Science such as traditional numerical libraries. Unless many familiar and widely used algorithms, libraries, and applications are substantially rethought and rewritten, they simply will not be able to exploit the power that new generations of multicore and hybrid processors offer. At stake are much higher degrees of concurrency and computing power than older architectures, unlike the traditional designs, they will also force software developers to take responsibility for managing all this new complexity in order to achieve significant performance increases. The workshop attendees will present work on software frameworks that dramatically simplify the software process on these designs which are still so new that extensive experimentation is needed to see how well old techniques apply and to find more flexible and adaptive programming models the new regime will require.
Submissions of high quality papers describing mature results or on-going work are invited. Topics for submission include but are not limited to:
We hope this workshop will invite open and exploratory discussions among the participants on the challenges of multicore hybrid architectures and their influence on numerically intensive computing.
| Submission Deadline: | April 17, 2007 |
| Notification: | May 17, 2007 |
| Final Copy Due: | June 10, 2007 |
Prepare your paper with free style no more than 15 pages (typically 4-8 pages using single space, 11-point font) in PDF format.
Submit your paper to
Accepted papers will be published by Lecture Note in Computer Science (LNCS), Springer (pending). At least one author of each accepted paper is requested to register and present the work at the conference; otherwise the paper will be removed from the digital library after the conference.
Distinguished papers, after further revisions, will be considered for a journal special issue.