Accuracy at Speed: How have Simulations become Faster

Reservoir simulation is computationally intensive, often solving millions of equations to model complex physics (e.g., multiphase flow, thermodynamics, geomechanics, etc). In the past, simulators ran on single CPUs and were limited by sequential processing speed. But today, the simulators leverage parallelism at multiple levels:

1. Distributed computing across many nodes (using MPI – Message Passing Interface)
2. Shared-memory multi-threading on multi-core CPUs (using OpenMP)
3. Acceleration via GPUs (Graphics Processing Units).

However, one often overlooked aspect is I/O and data handling. HPC improvements also involve faster reading/writing of large datasets and smarter compression to reduce communication overhead.

GPU Acceleration: A Game Changer in Simulation Speed

Perhaps the most exciting leap in simulation speed has come from GPUs. They offer a massively parallel architecture with thousands of cores that can perform many operations simultaneously. Initially developed for rendering graphics, GPUs have proven extremely effective for the linear algebra and vectorizable computations at the heart of reservoir simulation.

NVIDIA 's GPUs have become nearly synonymous with accelerated computing in many industries, and oil & gas is no exception. In fact, Computer Modelling Group announced a collaboration with NVIDIA to leverage the latest GPU technology for its simulators. By utilizing NVIDIA’s full-stack platform, including high-end H100 Tensor Core GPUs and the new GH200 Grace Hopper “superchip” (which pairs a GPU with a fast ARM-based CPU), we have unlocked substantial improvements in computational speed while maintaining our hallmark technical accuracy.

It’s worth noting that GPUs are not a silver bullet for every problem. Some parts of reservoir simulation (complex well management, certain physics) don’t parallelize easily on a GPU. Early claims that GPUs could replace CPUs entirely were met with some skepticism. In practice, the winning formula has been hybrid CPU+GPU computing. 

Advancements in CPU Technology (Including ARM)

While GPUs steal a lot of the spotlight, traditional CPU architecture has also advanced, offering new opportunities for speed gains. One notable trend is the rise of ARM-based CPUs in HPC. Historically, reservoir simulators were tuned for x86 processors (Intel/AMD), but ARM chips known for power-efficient performance are now reaching server-class capabilities.