Reservoir Modeling Performance Enhancements: How Real is the Buzz around GPU?

13 01 | 22

Before the turn of the 21st century, GPU (Graphics Processing Unit) computing was commonplace, but mostly used for computer gaming and professional graphics rendering. Now recent advances in technology have brought blazingly fast GPU computing to scores of industries, including oil and gas exploration and production.

The possibility of using GPU computing for reservoir simulation and modeling is an interesting topic – GPU’s vaunted high performance characteristics do sound promising for delivering fast reservoir simulations and enhancing overall performance. Or do they?

GPU by itself might be ideal if you only need to simulate uncomplicated black-oil reservoirs. And, if you don’t need to take aquifers into consideration. And, if your downhole equipment is fairly straightforward rather than sophisticated. And, if your surface networks aren’t all that complex. In those specific instances, GPU-only reservoir simulation runs might fill the bill. But be aware that, contrary to some assessments, GPU can’t speed up every simulation. In the real world, GPU all by itself can efficiently model only a limited subset of computer instructions; it isn’t equipped to handle every reservoir-modeling job. However, if you’re trying to make an impression by running academic, artificial benchmark tests such as SPE 10, GPU-assisted computing can leave its CPU-only competition in the dust!

But let’s take a look at reality. Your company has already made a hefty investment in the hardware you currently use for reservoir modeling – workstations, clusters, or perhaps an internal “cloud.” It is likely that this equipment is CPU-based, and it’s been doing a good job, but you wish it could run your simulations faster. So you’re tempted to follow the hype and give GPU-only reservoir simulation a try.

Be ready to spend the big bucks for dedicated and expensive hardware! And take into consideration you may have problems running your models due to the very real constraints that burden GPUs.

Investing in infrastructure for GPU-only reservoir simulations simply is not practical for E&P companies. This is because you need the ability to model a variety of different reservoir and production scenarios, both uncomplicated and highly complex. Limiting yourself to GPU-only processing means a both a loss in versatility plus the additional high cost of expensive hardware. So in reality, using GPU-only computing for reservoir simulations is “no way José.” But by judiciously adding GPU power to your existing CPU-dominant systems it’s possible to kick your reservoir modeling performance into high gear.

The Answer? The Practical Solution for Enhancing Speed and Performance Comes from Hybrid CPU+GPU Computing

It’s time to include GPU power in your reservoir modeling toolkit. After all, a CPU may have multiple cores that get work done quickly, but a GPU is built for multitasking and can have thousands of cores that focus on processing many small repetitive tasks. While GPUs are faster than CPUs, they are not as versatile. CPUs have extensive instruction sets so they can perform many more kinds of tasks than a GPU. When used appropriately, GPU can significantly lower the processing time of many reservoir simulations – particularly those that spend a lot of time in linear or nonlinear solvers.

For geoscientists, hybrid CPU+GPU means that computationally expensive geostatistical algorithms are sent to the GPU, where calculations can be accelerated more than four times. That speed makes a big difference in complex geostatistical simulation runs, frequently delivering same-day versus next-day results.

You’ll need to have a powerful integrated software program to reap all the benefits of hybrid CPU+GPU processing – software that’s capable of letting you fine-tune simulation runs by balancing how the available CPU and GPU is used for each model. You want to get maximum performance and speed. You need tNavigator®, the industry-leading high-performance tool for complex reservoir simulations.

tNavigator – Simulation Software Built for Hybrid CPU+GPU Environments

In addition to its excellent scalability with large models, the tNavigator software suite is a full-geoscience and full-physics approach to modern reservoir management. It consists of a suite of ten integrated modules, and it’s constantly evolving (four updates per year, every year!) so it adapts to any new technology quickly.

tNavigator incorporates hybrid CPU+GPU processing, so it can efficiently handle virtually any reservoir simulation task you throw at it. tNavigator’s GPU capability gives you four options to choose how the software load-balances between CPU and GPU processing:

  1. GPU on linear solver for all three types of simulation models – black oil, compositional and thermal;
  2. GPU on linear solver plus compositional flash option for black oil and compositional models;
  3. GPU on linear solver plus compositional flash plus relative permeabilities for black oil and compositional models;
  4. GPU on linear solver plus compositional flash plus relative permeabilities plus Newton solver for black oil models.

Hands down, tNavigator is preferred comprehensive environment for interpretation, static, and dynamic modeling using hybrid CPU+GPU architecture.

Speed + Flexibility + Functionality + Affordability = tNavigator

Even with its advanced functionality, tNavigator’s pricing is extremely competitive. Ask any of our users – cost is never the issue. Chances are you’ll pay much less for tNavigator’s hybrid CPU+GPU software and get much more value than with less robust offerings from competitors.

To learn more, and try tNavigator yourself, request a demo or visit the Contact Us page and call one of our 20 offices located throughout North America, South America, Europe, the Middle East, Africa, Russia, China, Southeast Asia and Australasia.

Related Articles: