The NOMAD mini-apps: A suite of kernels from ab initio electronic structure codes enabling co-design in...

Area of Science:

• Computational materials science
• High-Performance Computing (HPC)
• Quantum chemistry

Background:

• Electronic structure codes are crucial for materials simulation but computationally intensive.
• Optimizing these codes is essential for leveraging advanced High-Performance Computing (HPC) resources.
• Existing codes face challenges in efficiently utilizing emerging exascale architectures.

Purpose of the Study:

• To develop a suite of mini-applications for optimizing computational kernels in ab initio electronic structure codes.
• To facilitate the co-design of hardware and software for future exascale systems.
• To provide tools for profiling and benchmarking code performance on HPC systems.

Main Methods:

• Identification of key computational kernels contributing significantly to execution time in flagship applications.
• Development of a mini-application suite based on these kernels.
• Integration of a CMake build system for easy deployment on various HPC systems.
• Leveraging code from the NOMAD Center of Excellence (e.g., ELPA, exciting, Abinit, FHI-aims).

Main Results:

• A deployable suite of mini-applications targeting critical computational kernels.
• Demonstrated ease of compilation and execution on diverse HPC environments.
• Established a framework for profiling and benchmarking code performance.
• Provided insights for both software optimization and hardware design for exascale.

Conclusions:

• The mini-application suite enables focused optimization of electronic structure codes.
• Facilitates the co-design process for exascale computing in material science.
• Promotes more accurate and efficient simulations of novel materials.