Italian multinational Eni is putting its new HPC4 supercomputer to good use, using all 3,200 of the system’s GPUs to run 100,000 oil reservoir simulations in record time.
According to the Eni announcement, they employed a high-resolution model of a deep-water reservoir comprised of 5.7 million active cells and simulating 15 years of production. The idea was to determine how the hydrocarbons would behave when the reservoirs are drilled. In this case, the model was applied to 100,000 different geological instances, using a single NVIDIA Tesla P100 GPU per instance. The entire run across the 3,200-GPU HPC4 supercomputer was completed in just 15 hours.
This is not just a demonstration for bragging rights. Being able to model reservoirs at this speed and scale has real value to Eni’s business. From the announcement:
“This run is not only proof of computing power, but also the first practical step to provide all of Eni’s reservoir engineers access to a powerful processing tool for more accurately quantifying subsurface uncertainty and continually incorporating data into ‘live’ models on operating assets. It will further enhance Eni’s capability to accelerate the time-to-market of projects and deliver outstanding reservoir management strategies for all producing fields”
The exceptional throughput was the result of being able to complete each simulation on a GPU in around 28 minutes, a task that would normally take a few hours on a CPU. To extract the performance from the graphics processors, the Eni engineers used ECHELON, a GPU-optimized reservoir simulation code provided by Stone Ridge Technology.
“Stone Ridge was an early adopter of NVIDIA GPUs for high-performance computing and we built ECHELON from the ground up to exploit the technology.” said Vincent Natoli, founder and president of Stone Ridge. “We see the benefit of that choice now as ECHELON has become faster and more capable with each generation of NVIDIA product. We significantly outpace our CPU-based competitors in both performance and scalability and each year the gap has become more pronounced.”
At 18.6 petaflops, HPC4 became the world’s most powerful commercial supercomputer when it was installed four months ago. It also contains more GPUs than any other system used by the oil and gas industry, or any industry for that matter. When the Summit supercomputer at the Department of Energy’s Oak Ridge National Lab comes online later this year, it will be equipped with more than 25,000 GPUs – about eight times as many as in the Eni machine. And those Summit GPUs will be V100s, which deliver about 50 percent more peak flops than the HPC4’s P100s.
Alas, Summit is not likely to be running reservoir simulations for commercial concerns anytime soon. But if it did, this same 100,000-reservoir run could probably be dispatched in less than two hours. Since the fastest commercial supercomputers lag the performance of the top supercomputers at national labs by four or five years, oil and gas companies will probably not have a Summit-sized until 2022-2023.