Page 403 - Intelligent Digital Oil And Gas Fields

340 Intelligent Digital Oil and Gas Fields

sufficiently different to be called a quantum leap in modeling technology.
However, since these models are really designed to enable practical imple-
mentation of the closed-loop system and provide users with an interface to
collaborate and manage all important reservoir management decisions from
daily to long-term ones, it can indeed be considered a quantum leap in how
it enables near real-time data-driven reservoir management processes not
possible to traditional approaches.

9.6.3 High-Performance Computing for the Future DOF

For many industry experts, the advancement in high-performance comput-
ing (HPC) and cloud computing are seen as key enablers for the next-
generation DOF. The future is now and the technology vendors are
aggressively investing in graphic processing unit (GPU)-based reservoir sim-
ulation technology (e.g., Stone Ridge Technology (SRT), 2017 with its
ECHELON simulator and Rock Flow Dynamics (RFD), 2017 with its
tNavigator simulator) with the vision to transform the future DOF. How-
ever, as noted in the interview with Vasilii Shelkov, CEO at RFD, in the
past few years there has been very little progress in accelerating reservoir
simulations, both for shared memory (workstations, laptops) and distributed
memory (public and private clouds) systems.
The typical performance improvements expectations from a CPU
upgrade have diminished to 5–10%, which quickly became unacceptable
and forced the reservoir simulation community to look for new ways to
accelerate. Since performance of reservoir simulations is dominated largely
by throughput and latency of the memory system, the emergence of general
purpose GPU computing and its fast memory started to attract the attention
of software developers. As of now, more and more companies show that
CPU-GPU and GPU-GPU systems can easily challenge scalar performance
of classical CPU-CPU systems. The development of new CPU vector
processing tools like AVX512 could make compute-bound parts of reservoir
simulations more competitive with GPU, but will not help much with
memory-bound calculations. Using general purpose GPU computing for
reservoir simulations has already shown potential and with the merge with
CPU and memory systems may eventually become next-generation tech-
nology to boost reservoir simulation performance by an order of magnitude
or more in the next couple of years. In fact, the lack of competition is per-
ceived by Shelkov as the only major hindrance for progress of HPC and
cloud computing in the future of the DOF.