Since the last decade, radio astronomy has started a new era: the advent of the Square Kilometer Array (SKA), preceded by its pathfinders (like Low Frequency Array, LOFAR, or MeerKAT), will produce a huge amount of data that will be hard to process with a traditional approach. This means that the current state-of-the-art software for data reduction and imaging will have to be re-modelled to face such data challenge. In order to manage such an increase in data size and computational requirements, scientists need to exploit modern HPC architectures. In particular, heterogeneous systems, based on complex combinations of CPUs, accelerators, high-speed networks and composite storage devices need to be used in an efficient and effective way.

Our goal is to develop a software for radio imaging, that is currently one of the most computational demanding steps of the radio astronomy data processing, both in terms of memory request and computing time. The GPU porting is a key point that allows to make the most out of the accelerators parallel computational capabilities, minimizing the communication and data movement.

Starting from the original code presented in Gheller et al. (2023), I will present the implementation of the Fast Fourier Transform (FFT) on the GPUs adopting the distributed version of the NVIDIA optimized library (so-called cuFFTMp) to be able to allot the large datasets produced by the radio telescopes across multiple GPUs. This is a key point for the GPU development of the code, given that the size of the involved problems is so huge that cannot be handled by a single accelerator.
I will show the results in terms of speedup and scalability of this new accelerated version of the code based on a scientific case, namely real LOFAR VLBI data, and discuss the comparison with the CPU version of the FFT presented in the original code.

Overall, we would set a new way to approach not only radio astronomy, but astrophysical software all-round. This will represent the first example of radio imaging software enabled to GPUs, becoming a potential state-of-the-art work for the future SKA software suite.