astroCAMP-data-v1.0: SKA-Low visibility datasets and reference dirty images for cross-layer co-design benchmarking

astroCAMP-data-v1.0 This dataset accompanies the astroCAMP paper published at PASC26, a community benchmark for evaluating radio-interferometric imaging pipelines under performance, energy, scientific quality, and sustainability constraints. The framework is described in the associated publication (arXiv:2512.13591) and hosted at https://github.com/SEAMS-Project/astroCAMP. Context The Square Kilometre Array (SKA) will process petascale imaging workloads under strict power (1–5 MW) and cost envelopes. astroCAMP provides standardised datasets, reference outputs, and a unified metric suite to enable reproducible hardware–software co-design evaluation across heterogeneous architectures (CPU, GPU, FPGA, ASIC). This work is carried out within the SEAMS project (SNSF grant 220194, https://data.snf.ch/grants/grant/220194; ANR grant ANR-23-CE46-0010, https://anr.fr/Projet-ANR-23-CE46-0010), contributing to the design and optimisation of SKA Science Data Processors. Contents Visibility data (input workloads). Two SKA1-Low benchmark datasets simulated with OSKAR using the GLEAM source catalogue at 151 MHz, single-precision floating point, in CASA Measurement Set format version 2: oskar_skalow_gleam_64t_151MHz_64c_single.ms.tgz — 64 time steps, 64 channels (large workload, ~12.1 GB) oskar_skalow_gleam_8t_151MHz_8c_single.ms.tgz — 8 time steps, 8 channels (lightweight workload, ~205 MB) Reference dirty images (quality ground truth). WSClean output dirty images for each visibility dataset, 4096×4096 pixels at 17.578 arcsec/pixel, FITS format. Used as reference for astroCAMP algorithmic quality metrics (RMS, PSNR, dynamic range, astrometric and photometric error). Energy and performance traces. Anonymised power and performance counter measurements recorded during WSClean + IDG imaging runs on a KUMA node at EPFL SCITAS using an AMD EPYC 9334 CPU and an NVIDIA H100 GPU. No user-identifiable or host-identifiable data is present. These traces constitute the ground truth for astroCAMP system-level metrics: energy-to-solution, energy efficiency, carbon footprint, and cost per job. Three trace types are provided: .log: WSClean log including GPU/CPU performance and energy metrics for IDG as reported by PMT .pdu: Timestamped node-level PDU measurements in Watts .monit: Timestamped resource-monitoring records for a WSClean process, capturing e.g. per-GPU utilization, memory, power and temperature, CPU package power, process I/O counters, CPU usage, and RSS memory. How to use The provided measurement sets can be processed with WSClean to reproduce the results. License Creative Commons Attribution 4.0 International. The license allows re-distribution and re-use of a licensed work on the condition that the creator is appropriately credited. Cite as If you use this dataset, please cite it as: @dataset{astroCAMPdata2026, author = { Etienne Orliac, Denisa-Andreea Constantinescu, Rubén Rodríguez Álvarez, Jacques Morin, Mickaël Dardaillon, Sunrise Wang, Hugo Miomandre, Junior Mbuyi, Yves Lopes, Xavier Ouvrard, Miguel Peón-Quirós, Jean-François Nezan, David Atienza}, title = {astroCAMP-data-v1.0: SKA-Low visibility datasets and reference dirty images for cross-layer co-design benchmarking}, year = {2026}, publisher = {Zenodo}, version = {1.0}, doi = {https://doi.org/10.5281/zenodo.20093790}} together with its associated publication: @inproceedings{astroCAMPpasc26, authors = {Denisa-Andreea Constantinescu, Rubén Rodríguez Álvarez, Jacques Morin, Etienne Orliac, Mickaël Dardaillon, Sunrise Wang, Hugo Miomandre, Miguel Peón-Quirós, Jean-François Nezan, David Atienza}, title = {astroCAMP: A Co-design Analysis and Metrics Platform for SKA-scale Radio Interferometric Imaging}, booktitle = {Platform for Advanced Scientific Computing Conference (PASC ’26), June 29-July 01, 2026, Bern, Switzerland}, year = {2026}, doi = {https://doi.org/10.1145/3815572.3815738},}