High-performance computing has entered a period of rapid change. The emergence of multi-core and many-core architectures along with the increasing competitiveness of hardware accelerators such as General Purpose Graphical Processing Units (GPGPU) and Field Programmable Gate Arrays (FPGA) presents opportunities of rapid cost/performance gains.
However, it also creates great uncertainty on which, among all the possible modern HPC architectures, best fits the computational needs of Earth Sciences. Furthermore, taking full advantage of these architectures will require more than a simple adaptation of existing algorithms to new hardware; it will require the design of new numerical algorithms and the development of new programming models that enable fast and efficient implementation.
In the past few years, CEES has been successful in exploring the opportunities and challenges introduced by new architectures. It has collaborated with IT vendors by providing a clear description of important industry-specific methods that could be better supported by future hardware devices. CEES also has an ongoing collaboration with the Pervasive Parallelism Laboratory in the Computer Sciences Department at Stanford.
The goals of SESAAI are to:
- Evaluate modern High-Performance Computing (HPC) architectures for reservoir-simulation and seismic-imaging algorithms.
- Develop new algorithms that take advantage of modern HPC architectures.
- Develop "data streaming" abstractions that facilitate efficient porting of reservoir-simulation and seismic-imaging codes to modern HPC architectures.
- Influence the future technological offering by Information Technology (IT) companies to better meet needs of reservoir-simulation and seismic-imaging algorithms.