+65 6591 8608

Our blog

Home Our Blog Accelerating reservoir simulation with GPUs.

Thursday21 February 2019

Accelerating reservoir simulation with GPUs.

oil rig 2 bought
Accelerating reservoir simulation with NVIDIA GPU.

Get your results in hours vs days!

As large and complex computational problems become more and more common across numerous industry segments, the need of  computation by HPC (High Performance Computing) industry increases. However, traditional CPU technology is no longer capable of scaling in performance sufficiently to address this demand.

The parallel processing capability of the NVIDIA Kepler GPU allows it to divide complex computing tasks into thousands of smaller tasks that can be run concurrently. This enables computational scientists and researchers to address some of the world’s most challenging computational problems up to several times faster; and reservoir simulation could be one of those challenges.

What is the purpose of Reservoir Simulation?
Reservoir simulation models are used by oil and gas companies in the development of new fields. Also, models are used in developed fields where production forecasts are needed to help make right investment decisions. Reservoir Simulation has two main goals: Determine best locations and schedules for wells to optimize production and predict future production rate (from a reservoir model).

Why it’s so important?
Every energy company aims to make the best of their large drilling budgets. Drilling wells in the right place, and with accurate knowledge of the associated risks, really impacts the bottom line. When improved, accuracy leads more to the pay zone and less to uncertainty of the production process, which helps to increase the ROI significantly.

As building and maintaining a robust, reliable model of a field is often time-consuming and expensive; models are typically only constructed where large investment decisions are at stake (e.g. land wells can cost more than $10M each, when deep sea wells can cost more than $100M each). So when it comes to choosing the drill sites we need to be sure that that all the possible risks were reduced.

What are the challenges in Reservoir Simulation?
Reservoir simulation is an active field of research, which includes modelling and numerical discretization amongst its topics. When developing numerical software for reservoir simulation, one faces the challenge of solving large, sparse linear systems.

Nowadays some of the main challenges are:
• Optimizing production which requires fast turn-around for many “what-if” cases;
• Commercial simulators that could take days for 1M cell model;
• High-accuracy reservoir models can contain between 10M and 1B cells (and you might need hundreds of years for large scale simulation)!

How to solve these challenges?
Finding a solution to this may prove to be a bottleneck computation, where exploiting information about the problem at hand may be crucial for performance. Judging from our customer’s experience, when it comes to solving a linear system, over 95% of running time is purely spent on the solution of linear systems. So it is clear that if the linear systems are solved efficiently, the whole simulation could speed up dramatically.

And this is where Tesla GPU accelerator comes in useful.
GPU, which is usually used for graphics processing, is also powerful in float point calculation. Initially developed for 3D graphics rendering, GPUs have also become widely used for scientific computing over the last decade. With their massive parallelism, GPUs offer attractive computational capabilities, and have been used to speed up computations in a wide field of applications.

Our experience:
One of our customers in Oil&Gas used applications for Reverse Time Migration and Reservoir simulations. They were mostly interested in reducing costs and decreasing the risks of their projects (however they were not familiar with GPU-technology) thus they have asked us to create a Proof –of-concept (POC) for their projects in order for them to be able to evaluate potential performance results.

While developing the POC we used our cluster with the latest X8000 Exxtreme Server with 4 Tesla K20 cards (and an opportunity to increase the number of graphic cards up to x8 GPUs in 4U form-factor), FDR InfiniBand interconnect (56G) and Parallel File Storage (based on Lustre OS).

These are the end results that were received by our customer after 1 week of testing:
1. Their RTM-application received 15x-times increase in performance (3:15 hours instead of 49 hours of processing);
2. Simulations with more than one million particles exhibit 22x-times speedup over using a CPU solely.
3. GPU rendering delivers up to 40x acceleration (1 hour instead of 2 days of rendering), increasing the massive data crunching efficiency.
As a result users get a much more realistic rendering in 3D taking visualization to the next level, which helps them save their money as well as their time.

Bottom line is
Reservoir Simulation is a critical topic, where nothing can be left out. Avoiding overpressure zones, identifying fracture areas that could affect drilling all these factors should be taken into consideration, carefully processed and visualized with maximum details. Working efficiently to reduce project turnaround without compromising quality has a definite impact on customer performance. It also implies that the data volumes are constantly growing and with the users’ need to see more details the data processing becomes more and more compute intensive.

EXALIT has wide experience in implementing hardware platforms equipped with many NVIDIA® Quadro® and NVIDIA® Tesla® GPUs (including the latest, based on new Kepler architecture - K20, K40 and K80) that significantly increase the amount of data that can be analysed in a given timeframe and allow Seismic companies to improve subsurface understanding, reduce risk in oil and gas exploration and exploitation. Also our specialists have wide experience & expertise in porting scientific software solutions for GPU multi-core architecture. If you would like to find out more information or join one of our trainings, please feel free to contact us.

JA Minisite

2857.orig.q75.o0 - Copy IntelTPP amd  tesla preferred partner  quadro partner  qctlogo e   Mellanox APAC Partner   1