Enhancing EDF’s Fluid Dynamics Simulations with NVIDIA Nsight Profilers

Électricité de France (EDF), a leading multinational electric utility company, is advancing its computational fluid dynamics (CFD) simulations through collaboration with NVIDIA, leveraging the power of NVIDIA’s Nsight Profilers. This partnership aims to enhance EDF’s code_saturne application, an open-source tool developed in 1997 for simulating complex fluid dynamics flows, critical for power plant safety assessments and lifetime extensions.

Streamlining GPU Porting

The transition from CPU to GPU applications promises significant performance improvements, allowing for larger scale problem-solving at increased speeds. The process, while initially demanding, yields substantial throughput and efficiency benefits. NVIDIA’s suite of tools, particularly Nsight Systems and Nsight Compute, supports this transition by identifying acceleration opportunities and optimizing kernel performance.

EDF’s Collaborative Efforts

EDF, in collaboration with AWS and Aneo, is iteratively porting code_saturne to NVIDIA GPUs, enhancing its capability while maintaining its modular architecture. This effort is supported by AWS Cloud, providing scalability and accessibility. The project underscores the potential of NVIDIA platforms to accelerate complex simulations effectively.

Analyzing and Optimizing Code

Nsight Systems plays a crucial role in prioritizing code segments for porting by identifying bottlenecks. The use of CUDA managed memory facilitates seamless data migration between CPU and GPU, ensuring consistent code usability. Annotations through NVIDIA Tools Extension (NVTX) further enable detailed tracking and analysis of the porting process.

Figure 1 in the original source illustrates an Nsight Systems report, showcasing the iterative process of a code_saturne simulation, identifying areas for code optimization. This visual representation aids in pinpointing time-intensive routines, guiding developers in improving performance.

Identifying Porting Opportunities

Through detailed NVTX-annotated analysis, EDF identified CPU segments with potential for GPU porting. Addressing these segments reduced CPU-to-GPU memory transfers, minimizing idle GPU time. The result was an impressive 18x speedup for specific computations, as depicted in Figure 2 of the original source.

Ongoing efforts aim to further optimize GPU kernels, utilizing Nsight Compute for enhanced performance. This step is essential for maximizing the benefits of GPU acceleration across the entire application.

For more information, visit the NVIDIA Developer Blog.

Image source: Shutterstock