Oil and gas high performance computing at Rice (April 2014)

Around 500 hear from Nvidia, Aramco, Intel, HueSpace and SpiralGen on HPC applications.

The oil and gas high performance computing (HPC) workshop at Rice university is experiencing quasi Moore’s law growth in attendance, now at around 500. The ‘forum for HPC professionals’ explores HPC trends and challenges and sets out to develop science and engineering workforce pipeline for the industry.

Nvidia’s Bill Dally announced the end of ‘Dennard scaling’ as today’s microprocessors are energy limited and ‘not getting any faster.’ For Nvidia, the answer lies in ‘hybrid’ CPU/GPU computing with its poster child, the Oak Ridge Cray/Titan supercomputer with 18,688 Tesla GPU providing 90% of its 27 petaflop peak bandwidth. On the programming side ‘pervasive parallelism’ can be done the hard way or leveraging libraries like Nvidia’s Cuda. The cloud represents another way forward, offering a ‘1kW experience in a 1W handheld’ and simplifying data management.

Majdi Baddourah presented Saudi Aramco’s ever rising reservoir modeling cell count which stood at 700 million in 2013 and is headed for the giga scale model real soon now. Aramco is adapting industry standards for ‘unified data exchange’ with a tip of the hat to OpenSpirit and Resqml.

Tesla’s aren’t the only co-processors in town. Leonardo Borges presented work done with Petrobras using Intel’s Phi architecture for reverse time migration. Intel’s unified programming model means that the same code is used across the cpu and the coprocessor (read no need for Cuda) and the dual IvyBridge/quad Phi combos showed good scalability and 23.7 GSample/s for a 5.5 GB domain

Michele Isernia presented HueSpace’s ‘next generation’ development platform for E&P visual computing that sets out to bridge the ‘major technology gap’ between current E&P software and the latest HPC technologies. Hue’s demo ran on a 36 teraflop Super Micro X9DAI hybrid, cloud-enabled with Calgary Scientific’s PureWeb technology.

Brian Duff presented SpiralGen’s technology for automatic generation of 3D fast Fourier transforms. Spiral is ‘software that writes software’ that is parallelized and optimized for various target architectures, in this case an IBM Blue Gene/Q system . The company is backed by the Argonne National Laboratory. More from SpiralGen. 

Click here to comment on this article

If your browser does not work with the MailTo button, send mail to info@oilit.com with OilIT_1404_10 as the subject. Web use only - not for intranet/corporate use. Copyright 2014 The Data Room - all rights reserved.