More Oil, Less Toil: How GPUs Can Make the Most of Fossil Fuel Resources

Talk about a space-saving device.

GPUs have shrunk half a football field’s worth of computing power into less space than a ping pong table on their way to helping the oil industry reach a performance milestone – simulating and oil reservoir with a billion computational cells

They did it in a fraction of the time that CPUs require, while using  fewer computing resources and 10 percent of the power.

Oil companies model underground reservoirs before they drill to figure out how to extract the most petroleum with the least financial and environmental risk. A billion-cell simulation is extremely challenging because of its complexity; cells represent different characteristics of the reservoir including the soil and fluid pressure.

“Few companies have done it, and no one has done it faster,” said Vincent Natoli, CEO of Stone Ridge Technology, maker of the ECHELON petroleum reservoir simulation software. Unlike competing products, the software runs all its calculations on GPUs.

The Football Field and the Ping Pong Table

Stone Ridge completed its simulation of a billion computing cells in 92 minutes using 30 IBM OpenPower servers equipped with 120 NVIDIA Tesla P100 GPU accelerators. They used it to estimate 45 years of production from 1,000 wells in a model based on giant reservoirs found in the Middle East. The larger the number of computing cells, the more engineers can learn about reservoirs.

A model of a reservoir simulation using a billion computing cells.
A model of a reservoir simulation using a billion computing cells. Animation courtesy of Stone Ridge Technology.

“The GPU calculation speed lets reservoir engineers run more models and ‘what-if’ scenarios than previously,” Natoli said. “That way they can produce more efficiently, lowering costs and making more responsible use of limited resources.”

Previous energy industry attempts to simulate a billion computational cells with CPUs required hundreds or even thousands of servers and about a day of computing time. One recent effort used more than 700,000 processors in a server installation that occupies nearly half a football field, according to Natoli.

Stone Ridge’s GPU-accelerated calculation needed only two racks of IBM machines, which fits into less space than ping pong table.

High Performance Computing for All

Energy companies already rely on GPU computing to figure out where to find reserves. With GPU-accelerated reservoir simulations, they can now determine how to drill to efficiently make the most of underground resources.

Although most oil companies don’t need to model a billion cells, even smaller simulations often demand a great deal of CPU computing resources. Using GPUs, ECHELON can run the more common simulation of 8 million cells on just one of our Tesla P100 accelerators in a desktop workstation.

“Most energy companies don’t have access to hundreds or thousands of computers,” Natoli said. “This democratizes high performance computing and puts it within the reach of more businesses.”

For more information, see the related press release or Natoli’s blog post.

To learn more about how AI computing is changing oil and gas and other industries, join us at the GPU Technology Conference, May 8-11, in Silicon Valley. Register for the conference at our GTC registration page.

Similar Stories