2025 Energy HPC Conference

​Click here to view the slides.
Click here to view the recording​​​.

A Pragmatic Approach to Optimize Execution Time and Cost of Complex Coupled-Physics Codes in Chevron’s HPC

Abstract: This work introduces pragmatic approaches for the systematic wall-clock time and execution cost optimization of complex codes, such as GEOS, in Chevron's Azure HPC environment. The target codes partition the computation at process and thread levels, need to scale to O(1000) of cores or O(100) of accelerators and run with minimal wall-clock times or cost on a diverse variety of processors and h/w platforms. We demonstrate that the performance of these codes is not a monotonically increasing function of the level of h/w resources they use, it varies with simulation model and, it is not easily assessed without running the code on specific h/w. Our approach relies on application profiling to identify Run-Time Configuration (RTC) space points (H,n_nodes,N_thr,n_{thr-rank},…) with minimal wall-clock time or cost and generate strong or weak scalability curves for each interesting simulation model. It leverages target h/w information to optimally place ranks and threads and to reduce the set of RTC points to assess, and it further “compresses” the profiling information to the optimal RTC for each specific node count. Here, n_nodes is the number of model “H” nodes, N_thr the total number of application threads, n_{thr- rank} the threads/rank and, “…” additional parameters like compiler optimization options. The profiling information among other includes initialization, linear-solve, non-linear implicit steps, and MPI times. We identify the performance of the linear and non-linear solvers with the profiling data at the best RTC point, and we gauge actual improvements as algorithms changes by SMEs. We have implemented this approach in a semi-automated run-time optimization framework. We demonstrate the ability of our methodologies to attain significant wall-clock time or cost savings results using GEOS and actual physical models. GEOS is an exascale-grade, multi-physics, multi-scale, simulation framework that advances the state-of-the-art in complex numerical analysis topics. Among others, it can simulate coupled flow, geomechanics and fracture models, including CO2 sequestration and storage, with simulation horizons of O(1000) of years.

Speaker: Michael Thomadakis, PhD - Senior Innovation and HPC R&D, Chevron Technology Center

Bio: Michael E. Thomadakis, after spending 3 ½ years at the Computer Science Department of Texas A&M University as post-doctoral and teaching faculty developing systems courses, joined the HPC Research Center at the same University where he led the design and implementation of a wide variety of supercomputer systems and carried out system and HPC application performance analysis and optimization. He subsequently joined the R&D division of the Shell Information Technology International where he evaluated, developed, and introduced innovative pre-GA technologies (Intel KNC, KNL, OmniPath, Nvidia GPUs, IB, etc.) to the HPC ecosystem. Subsequently he joined Mellanox Inc. where he analyzed and optimized the performance of several parallel distributed applications over different MPI stacks on IB fabrics. Michael is currently a senior member of the Innovation and HPC R&D division of Chevron where he is evaluating next-generation h/w and s/w technologies, optimizes parallel applications on a diverse set of h/w platforms, and is currently focusing on state-of-the-art, exascale-grade multi-physics HPC codes.

Authors: Michael Thomadakis (Chevron Technology Center), Pavel Tomin (Chevron Technology Center), Alex Loddoch (Chevron Technology Center) and Victor Magri (Lawrence Livermore National Lab, Hypre Project)

​Click here to view the recording​​​.

PGAS-Based Distributed OpenMP (DiOMP) for Seismic Modeling with Extension to GPU Computing

Abstract: We presented DiOMP in [1], but in this contribution we extend this PGAS-based OpenMP distributed implementation to supports OpenMP target offloading for GPU computing. By integrating the LLVM compiler, GASNet-EX library, and corresponding memory allocation for efficient GPU memory management, DiOMP simplifies programming compared to MPI+OpenMP, while maintaining competitive performance. Evaluation with kernels and an application demonstrates DiOMP’s scalability and productivity for heterogeneous systems.

Speaker: Mauricio Araya-Polo - Senior R&D Manager HPC and ML, TotalEnergies EP Research and Technology USA

Authors: Baodi Shan (SUNY Stony Brook), Barbara Chapman (SUNY Stony Brook), and Mauricio Araya-Polo (TE EP R&T US)

​Click here to view the recording
​​​
Optimizing the Delivery of High-Performance Workstations for Geophysical Workflows in Subsurface Exploration

Abstract: The energy industry is increasingly reliant on high-performance computing (HPC) and advanced digital tools to drive innovation in subsurface exploration and reservoir analysis. The design of IT environments typically involve a hybrid approach, combining on-premises infrastructure with cloud resources to meet the high demands of upstream workflows. Hybrid models can introduce several challenges, particularly in managing user access, coordinating virtual workstation environments, and ensuring the efficiency and performance of end-user experiences. Balancing workloads between on-premises and cloud platforms, while maintaining seamless access and optimal performance, requires careful orchestration to avoid bottlenecks and ensure that computational resources are effectively utilized. This session will highlight a case study of how Chevron revolutionized their IT infrastructure to optimize the delivery of high-performance workstations used for geophysical analysis. Chevron was able to replace outdated systems, streamline workflows, and unlock significant improvements in computational performance and end-user experience. Their modern hybrid environment spans both on-premise and cloud resources, dynamically provisioning and controlling virtual workstations in the cloud for efficient compute usage without compromising performance. With their new solution, Chevron achieved a more efficient use of cloud resources, reducing operational costs by automating power management, provisioning, and virtual workstation allocation. This ensured that only the necessary computing power was deployed at any given time, enabling substantial cost savings while still maintaining the high performance required for demanding geophysical simulations and reservoir modeling. As a result, Chevron’s geophysicists were able to work faster, solve more complex problems, and increase their overall productivity. Key takeaways include: 1) the simplification of end-user experience; 2) the optimization of system management; 3) the adoption of modular, scalable solutions to ensure flexibility and adaptability, reducing the risk of disruption when technologies reach end of life.

Speakers: 
- Stephen Rigler - Senior HPC Cloud Engineer, Chevron
- Blake Ray - HPC Cloud Engineer, Chevron
- Karen Gondoly, MS - CEO, Leostream

Bios:
Karen Gondoly is CEO and VP of Product Management of Leostream Corporation. She has worked closely with IT decision makers across all major industries to help transform complicated deployments into highly scalable, performant, and automated hosted workstation environments.

Authors: Stephen Rigler (Chevron Corporation), Blake Ray (Chevron Corporation) and Karen Gondoly (Leostream Corporation)

Click here to view the slides.
Click here to view the recording​​.

Abstract: Accurately capturing topographic effects in wave-equation based seismic imaging algorithms such as RTM and FWI enhances their kinematic accuracy in land settings, improving imaging quality1. By representing free surface topography as an immersed boundary, complex geometries can be accurately captured whilst retaining use of cartesian, structured FD methods2. Eschewing curvilinear or unstructured grids, immersed boundaries are readily compatible with existing propagators and overarching imaging frameworks. However, devising a suitable boundary treatment and implementing the associated routines in the underlying kernel represents a substantial effort, whilst introducing additional complexity and potential technical debt.

To date, immersed boundary implementations, particularly in seismic contexts, have focussed on specific equations and boundary conditions. This specificity hinders extension to new equations, particularly when application-specific approximations are used. Furthermore, because these implementations rely on low-level interventions within the kernel itself, the cost of implementing such treatments rapidly becomes a barrier to practical deployment.

Constructing a framework which leverages symbolic computation to generate suitable immersed boundary treatments from a high-level specification of boundary geometry and conditions introduces a layer of abstraction between the user and underlying numerics. By avoiding major modifications to the underlying numerical methods and encapsulating the boundary treatment within a powerful abstraction, topography can be treated as a module of the propagator framework, avoiding the need for fundamental or extensive reworking to include topography. This facilitates topography implementation in overarching imaging and inversion workflows, whilst enabling code reuse across applications. Integration with Devito3, a DSL and compiler for stencil computations further reduces barriers to entry.

Using this framework, FWI workflows can be straightforwardly implemented, and we have successfully constructed tomographic gradients in settings featuring over a kilometre of irregular topographic variation. Furthermore, it was found that a topographic free surface results in improved illumination balance over that observed in a comparable flat-surface case or where topography is implemented as a damping surface.

By managing complexity through abstraction layers, whilst enabling generalisation of immersed boundary treatments across a wide range of physics, the approach developed represents a powerful means of handling the emerging challenge of topography in land seismic imaging.

Speaker: Edward Caunt, PhD - Research Scientist, Devito Codes

Bio: Dr Edward Caunt is a Research Scientist at Devito Codes, developing novel abstractions for finite-difference methods with a focus on seismic modelling and imaging applications. His PhD thesis explored the development of a generalised mathematical approach to immersed boundary implementation, enabling the automatic generation of numerical treatments for complex topography across a wide range of seismic applications. Applications of this work have been published in Geophysics, alongside presentations at numerous academic and industry conferences. His ongoing research and development work focuses on domain-specific languages (DSLs) and code generation for high-performance, high-productivity geophysical model development.

Authors: Edward Caunt( Devito Codes), Rhodri Nelson (Imperial College London), Fabio Luporini (Devito Codes), Mathias Louboutin (Devito Codes), Gerard Gorman (Imperial College London)