To address the challenge of cost-effectively profiling oil and gas deposits, an innovative Russian geophysical service company, JSC Gradient, located in Kazan, Republic of Tatarstan, turned both to the Institute of System Research of the Russian Academy of Sciences and to AMD to assist in the effort. This successful collaboration brought together JSC Gradient’s methodology of mathematical modeling of seismic wave propagation in solid media with one of Russia’s leading centers of High-Performance Computing research.

 

Historically, JSC Gradient started out with limited resources for using general-purpose processors to scientifically verify their methodology.

 

“Our customers, such as Gazprom, Tatneft and Lukoil, are constantly developing areas with increasingly complex geological structures,” said Evgeny Biryaltsev, LLC Gradient Technology and R&D Director of JSC Gradient.  “This demands solving the seismic wave propagation problem not only in the simplest 2D case, but also in 3D. In order to meet the computational demands we had to urgently upgrade our cluster.”

 

1.jpg

 

JSC Gradient consulted with experts at the Institute of System Research of the Russian Academy of Sciences to further improve the implementation. The Institute recommended a state-of-the-art professional solution from AMD -- AMD FirePro™ W9100 GPUs -- as the basis for the company's compute cluster, since the compute optimizations required both high memory bandwidth and as much GPU memory as possible.

 

“The AMD FirePro W9100 GPU algorithm benchmarks showed that the GPUs had superior performance for our needs,” said Dr. Denis Demidov, senior researcher at the Institute of System Research. “We approximated that 20 GPUs would be enough to solve the models of up to one billion elements, which was a requirement by JSC Gradient. The final design of the compute cluster included six nodes with four GPUs each to provide some reserve.”

 

2.jpg

 

After initial setup and tuning, conducted jointly by JSC Gradient, Institute of System Research of RAS, and LLC Baiko, the cluster was tested with a well-known LINPACK benchmark. The six-node cluster with 24 AMD FirePro W9100 GPUs showed impressive performance of 36.6 TFLOPs, securing 38th position in the Russian top 50 supercomputers rating. The cluster showed both the highest performance per compute unit and the best GFLOPS/Watt ratio on the list. This was also the highest ranked system in the list based on AMD FirePro GPUs.

 

3.png

 

“Currently, we are engaged in a number of major projects, in particular on the additional exploration of OAO Gazprom licensed areas in the Caspian region and Eastern Siberia and in projects in Latin America,” said Nikolay Shabalin Director of JSC Gradient. “All of these projects are characterized by complex geology, and we have to carry out large-scale numerical experiments on three-dimensional models. Powered by AMD FirePro GPUs, we are able to successfully cope with the new challenges and adjust our technology to further increase the success of drilling both in traditional fields and in the fields of shale hydrocarbons. The increased drilling success is something in which all our customers have interest in the face of declining hydrocarbons prices.”

 

AMD FirePro professional cards include the AMD FirePro W-series boards with active cooling for Workstations and the AMD FirePro S-series with passive heat sinks for Server integration. AMD professional graphics cards offer powerful performance for research where high performance computing is a critical factor, especially when it comes to locating critical resources such as oil and gas deposits.

 

JC Baratault is a Senior Development Manager for HPC GPU computing at AMD. Links to third party sites are provided for convenience and unless explicitly stated, AMD is not responsible for the contents of such linked sites and no endorsement is implied.

 

AMD, the AMD Arrow logo and FirePro are trademarks of Advanced Micro Devices, Inc. Other names are for informational purposes only and may be trademarks of their respective owners.