Novel acceleration technology is delivering a discontinuous jump of over one order of magnitude in compute capability to the earth sciences community. For example, Chevron announced a Maxeler system at the recent SEG meeting offering over 200x speedup compared to a single state-of-the-art Intel Xeon CPU core.
In particular, acceleration increases with problem size, suggesting that larger problems get more attractive than their smaller counterparts. Another component lies with the computation-to-communication ratio, where more computation, larger stencils in convolutions, and more complicated cases such as elastic modeling, accelerate much more than the simple case. Thus, the challenge lies in identifying the impact on seismic algorithm selection and design, or in other words: acceleration changes ‘what is fast’, offering new potential for groundbreaking earth sciences algorithm research.
Dr Oskar Mencer will be discussing the potential for acceleration at the Stanford Center for Computational Earth and Environment Science in December 2008. Dr Mencer is CEO of Maxeler Technologies and Consulting Professor at Stanford CEES.