IT-Systemarchitekt*in

Installing scientific software on a supercomputer can take hours — and that's if it works the first time. Kenneth Hoste has spent over a decade trying to make that problem disappear, and his answer…

NVIDIA cuPhoton is a new toolkit designed to accelerate scientific image analysis, particularly for astronomy and other fields requiring high-performance computing. The technical blog post details how cuPhoton leverages GPU acceleration to process astronomical images efficiently, focusing on two key tasks: image subtraction and dipole fitting for moving objects like asteroids. The toolkit uses a Gaussian-polynomial kernel basis for optimal image subtraction, ensuring flux conservation and compact kernel representation. For dipole fitting, cuPhoton employs a GaussianDipoleModel to identify moving objects through paired positive and negative flux regions in difference images. The blog demonstrates a synthetic example with a planted mover (simulated asteroid) to validate the workflow, showing how the toolkit can accurately recover the positions of moving objects. The post also covers scaling the workflow across multiple GPUs and nodes, making it suitable for large-scale astronomical surveys like those from the Vera C. Rubin Observatory. The toolkit is built on NVIDIA's CUDA ecosystem, with support for CuPy, Numba-CUDA, and CPU backends, and includes tools for candidate review and classification. The blog emphasizes the importance of proper image registration and kernel basis selection to avoid artifacts like bright rings or negative bowls in the residual images. It also provides practical examples and code snippets for implementing the toolkit, highlighting its potential to significantly reduce processing times for large astronomical datasets. The post concludes with a call to action for developers to explore the cuPhoton documentation and examples, positioning it as a powerful tool for scientific computing in astronomy and beyond.

Company Description: The Organisation for Economic Co-operation and Development (OECD) is an international organisation comprised of 38 member countries, that works to build better policies for better...

Julia language keeps AI honest in engineering

[EuroPython 2026 - S3B on 2026-07-17] 🎤 *Building your DSL compiler in Python by Mateusz Sokół* 🔗 https://ep2026.europython.eu/session/building-your-dsl-compiler-in-python 📝 Abstract: A…

MAPI is a modular web-based framework designed to unite legacy analysis code with heterogeneous backends (edge, on-premises, high-performance computing, cloud), dramatically accelerating data analysis workflows and offering a scalable flexible blueprint for modernizing synchrotron data analysis.

Scientific computing augmentation paving the way for the advent of next-generation synchrotron imaging research is discussed.

YouTube video by PyData

YouTube video by PyData

Reimagining how you share, collaborate, and run Jupyter notebooks — all in your browser.

Ternary and quaternary amorphous oxide semiconductors have many properties that make them promising candidates for use in electronic applications like display, memory, and back end of line logic. Howe...

Our PyTorch 2.8 Live Q&A webinar will focus on PyTorch packaging, exploring the release of wheel variant support as a new experimental feature in the 2.8 release. This feature is designed to improve…