ssc-cdi: A Memory-Efficient, Multi-GPU Package for Ptychography with Extreme Data

Area of Science:

• Computational physics
• Scientific software development
• X-ray imaging

Background:

• Ptychography reconstruction requires significant computational resources, especially for large datasets from modern X-ray facilities.
• Existing software packages may not always meet the specific needs of individual beamlines or user communities.

Purpose of the Study:

• Introduce ssc-cdi, an open-source software package for memory-efficient, single-node multi-GPU ptychography reconstruction.
• To provide a robust and customizable solution for ptychography data analysis, particularly for the Brazilian Synchrotron Light Laboratory (LNLS).

Main Methods:

• Developed ssc-cdi with reconstruction engines in Python (3.9.2) and C++/CUDA.
• Implemented memory-efficient algorithms for single-node multi-GPU processing.
• Benchmarked the software package using beamline data.

Main Results:

• ssc-cdi effectively reconstructs ptychographic data from beamline experiments.
• The software demonstrates efficient handling of extreme datasets without significant performance compromise.
• Benchmarks confirm the package's viability for high-resolution imaging applications.

Conclusions:

• ssc-cdi is a valuable, open-source tool for ptychography reconstruction, complementing existing community packages.
• It facilitates local expertise development and customized solutions for synchrotron light facilities.
• The software provides a robust approach for advanced, high-resolution imaging in X-ray science.