Algorithms and Applications for Emerging Architectures

 

To out-compute is to out-compete, in terms of accuracy (resolution), fidelity, complexity, certainty, turnaround time, and providing guidance for experiments. The ECRC enables diverse campaigns in science and engineering across KAUST and internationally to migrate effectively (in terms of human facility) and efficiently (in terms of hardware utilization and power consumption) to continually expanding computing performance. “Diverse” refers both to domain subject matter, from seismic imaging to genome-wide association studies, and to technique, from simulation based on first- principles models to machine learning where first-principles models are not known or less predictive per unit expenditure.

The ECRC fulfills this vision by being a source of software that is widely used for the implementation of efficient algorithms for analytics and simulation on emerging hardware. Common to these diverse investigations are the pressures from communication-austere computer architectures to exploit “data sparsity,” to exploit instruction-level concurrency, and to reduce synchronization. We also carry out some of our own leading-edge computational science and engineering. In terms of technology translation, we engage both with computer vendors (e.g., Cray, NVIDIA) and industrial users (e.g., Aramco, McLaren). We also train students and post-docs for the rapidly expanding workforce in simulation and big data analytics, internationally and in the Kingdom.

The ECRC vision fits in the “digital pillar” of KAUST’s strategic plan and it supports the other pillars to varying extents, especially energy. Approximately half of KAUST faculty engage in supercomputing and KAUST operates the most powerful supercomputer within 102 degrees of longitude and 113 degrees of latitude. An almost hemispherical leader in hardware should be a leader in software and algorithmic development for return on investment. In exploiting data sparsity in linear algebra on GPUs – a critical technology in spatial statistics and engineering optimization – and in rapid mesh traversal on many-core shared-memory accelerators – a critical technology in seismic wave propagation – ECRC is a global leader. This is evidenced by the incorporation of ECRC software into the offerings of NVIDIA, Intel, and Cray. We envision that every scientific accelerator chip sold will run KAUST software. While ECRC software will be part of the mostly invisible infrastructure of global HPC, many KAUST-authored applications that are built upon that infrastructure – some from ECRC sponsorship and some from beyond – will be highly visible because of their impact, from finding petroleum to moving an economy beyond petroleum.

In short, the ECRC envisions making KAUST a “Mecca” for high performance computing (HPC). The ECRC will become a destination for HPC researchers by becoming a source.