Petrobras Leverages NVIDIA Grace CPU for Enhanced Reservoir Simulation
Rongchai Wang
Sep 26, 2024 11:05
Petrobras achieves significant improvements in reservoir simulation using NVIDIA Grace CPU, with up to 4.5x faster solutions and 4.3x better energy efficiency.
Petrobras, a leading Brazilian energy company, has achieved remarkable advancements in reservoir simulation by utilizing the NVIDIA Grace CPU, according to the NVIDIA Technical Blog. The collaboration has resulted in up to 4.5x faster time-to-solution, 4.3x greater energy efficiency, and 1.5x higher scalability compared to traditional x86-based CPUs.
Enhanced Reservoir Simulation
Reservoir simulation is essential for optimizing resource exploration and production in the energy sector. Petrobras, known for its ultra-deepwater exploration, has leveraged high-performance computing (HPC) to minimize exploration uncertainties and improve success rates. The integration of the NVIDIA Grace CPU into their simulation workflows has proven to be a game-changer.
SolverBR Project
Solving linear systems, which can consume up to 70% of computational time, is a critical aspect of reservoir simulations. Petrobras collaborated with UFRJ and other research institutes to develop SolverBR, a CPU-based linear equations solver. This project aims to optimize sparse linear solvers for better performance and accuracy, integrating them into both proprietary and third-party simulators.
Transition to Arm-Based Architecture
Petrobras, in partnership with NVIDIA and Brazil’s CESAR Innovation Center, has successfully ported SolverBR to the NVIDIA Grace CPU, an Arm-based processor. The NVIDIA Grace CPU features a 72-core Arm Neoverse V2 and high-bandwidth LPDDR5X memory, delivering exceptional performance and energy efficiency.
Initial benchmarks demonstrated significant improvements, with the NVIDIA Grace CPU outperforming x86-based processors. This success was facilitated by robust open-source compilers and debuggers, enabling a seamless transition from x86 to Arm.
Performance and Energy Efficiency
Singularity containers were used to ensure reproducibility across different platforms. The NVIDIA Grace CPU showed superior performance, achieving up to 4.5x speedup over Petrobras’ baseline Intel Xeon Gold 6248 processor and up to 4.3x higher energy efficiency. These results highlight the advantages of the NVIDIA Grace architecture, including high memory bandwidth and advanced CPU Scalable Coherent Fabric.
Future Plans
Following these promising results, Petrobras plans to port and benchmark their entire geomechanical and reservoir simulation workflows to the NVIDIA Grace architecture. This initiative aims to further enhance their operational efficiency and reduce time-to-solution.
To explore more about the NVIDIA Grace CPU, visit the official NVIDIA page.
Image source: Shutterstock