Since 1986 - Covering the Fastest Computers in the World and the People Who Run Them

Language Flags
May 16, 2014

Linear Solver Library Grows Up

Tiffany Trader
NVIDIA OilReservoir 400x

NVIDIA’s AmgX is a linear solver library for large-scale industrial applications that entered beta production in January. Since then the software has matured, and Joe Eaton, lead for the AmgX product team at NVIDIA, wrote all about it in a recent blog. Eaton explains that V1.0 comes with classical AMG multi-GPU support and greatly improved scalability. The company is also releasing some performance numbers to back up its claims.

Eaton discusses a specific class of problem that has challenged the oil and gas industry, namely reservoir simulation, which is used to predict the behavior of wells with hydrocarbon deposits, as well as shale gas or shale oil fields. The associated models need to demonstrate flow through porous media, combined with flow through networks of fractures, piping and processing equipment. The media, however, is the most important element.

“Oil and gas deposits aren’t like big caves with lakes of oil,” writes Eaton, “they are more like complex, many-layered sponges, each with different pore sizes, stiffness and hydrocarbon content.”

He continues: “The flow is dominated by the interaction of pressure forces with the local and global characteristics of the media, and worst of all, we really don’t even know the exact configuration. There is no way to measure the properties of the rocks except by drilling cores, and bringing samples back to the surface. This sampling process gives some isolated data points on huge expanses of rock, and we have to interpolate or guess as to what lies between those ‘good’ data points by interpreting seismic data.”

Engineers build precise models with the available data, but even the best solvers have faced limitations, according to Eaton. For years one of the primary standard benchmarks used for reservoir simulations has been the Society of Petroleum Engineers 10th comparative simulation model, or SPE10 for short. Eaton writes that previously the AMG approach fell short of this benchmark, but not anymore. “Now we can show some great performance on this challenging benchmark, and start to engage with the reservoir simulation community, hydrology and water resources, and anyone interested in high fidelity geological models and flow through reservoirs,” he maintains.

In recent benchmark testing, NVIDIA tried out a 5 million cell version of SPE10 on a single K40 GPU with 2×10 core Intel “Ivy Bridge” E5-2690 CPUs. They also tested 10 million cells on a single GPU, and then did multi-GPU tests keeping the number of cells per GPU at least 5 Million. The results showed a 2x speedup on the expensive classical AMG setup phase, and a 4x to 6x speedup on the solve phase, which is where classical AMG works best.

Check out the post to read about the other interesting findings that came out of the testing.