DEFENSE

The Stanford Center for Computational Earth and Environmental Science (CEES) and Maxeler Technologies Inc. have agreed to collaborate in a research partnership to explore the opportunity of accelerating seismic computations - a critical advance in the quest to expand capacity for new interdisciplinary Earth science research. “We are combining CEES’s expertise and experience with seismic processing and Maxeler Technologies’ capabilities in acceleration of compute intensive applications” said CEES Director Jerry Harris, Professor in the Geophysics Department at Stanford. “It is only through close cooperation between seismic research and computer engineering that we can provide computing speed to meet the increasing demands of oil and gas exploration." From the Maxeler perspective, “by working with CEES we get real insight into algorithms and needs of Geophysicists which enables us to apply our acceleration technology to real problems” said Michael Flynn, Chairman of Maxeler Technologies and Emeritus Professor at the Stanford Computer Systems Laboratory, “Access to this outstanding center provides invaluable input for our acceleration specialists.” A wide range of advanced seismic algorithms are currently beyond reach of even very large computer systems. For example some of today’s high end imaging algorithms has been shown to produce 3-D images that lead to greatly improved results but they incur prohibitive computational costs. Research to make advanced algorithms computationally tractable is imperative for the seismic industry. Many of today’s and tomorrow’s applications rely on a relatively small subset of core routines. By accelerating these routines exciting new research opportunities are opened up for both academic and industrial researchers. On the other side, Maxeler's acceleration technology changes the game of high performance computing. Using an ordinary compute server, and plugging-in a Maxeler board, combined with applications-specific board configurations, Maxeler can accelerate application performance by over 10 times. To achieve such cost-efficient speedup, Maxeler uses a variety of hardware approaches including specialized chips such as FPGAs, in the acceleration boards. Furthermore, programming modern multi-core and multi-processor systems requires the user to rewrite applications and performance remains limited by the system’s memory bandwidth. Using custom accelerator boards provides an alternative to simply modifying software for multi-core use. Maxeler's accelerator boards have the potential of much higher performance by using parallelism that is tailored to a particular application. The Maxeler solution enables the use of acceleration technology within the normal software development process and business environment. The Stanford-Maxeler collaboration goes beyond core hardware acceleration by looking at the complete system including software restructuring, partitioning and using the specialized Maxeler plug-in board, while maintaining application flexibility. CEES is a research partnership created by the Stanford School of Earth Sciences, affiliates from the Stanford Computer Systems Laboratory and private industry. The goals of CEES include expanding the capacity for interdisciplinary Earth science research and engaging computer architects to design hardware and software systems that are better suited to Earth and environmental science problems.