Researchers have developed SACRA-K, a new numerical relativity code designed to simulate extreme astrophysical events such as the merger of black holes and neutron stars. This code, an adaptation to C++ with the Kokkos library from the previous Fortran code SACRA-MPI, retains the original physics and numerical methods, including BSSN spacetime evolution with Z4c constraint propagation and Berger-Oliger adaptive mesh refinement. The main novelty of SACRA-K is its "performance portability" across different hardware architectures, allowing it to leverage the power of graphics processing units (GPUs) and accelerated processing units (APUs).

SACRA-K was validated by comparing it against SACRA-MPI across various configurations, such as binary black hole systems, black hole-neutron star systems, and binary neutron star systems. The results show that discrepancies in the generated gravitational waveforms are well below the variability observed among independent codes and resolution-dependent variations within a given code. Furthermore, these differences remain at or below the distinguishability threshold of current gravitational-wave detectors. The code also preserves π symmetry at the bitwise level and exhibits second-order convergence in the gravitational wave phase during neutron star mergers.

In the smallest test configurations, SACRA-K proved to be approximately an order of magnitude faster on NVIDIA A100 GPU clusters or AMD MI300A APU clusters than Fortran SACRA-MPI on CPU clusters. The team has successfully scaled SACRA-K's performance up to 256 accelerator devices, highlighting its ability to exploit the massive parallelism of modern architectures. This advancement is crucial for numerical astrophysics, enabling faster and more efficient simulations of gravitational-wave-generating phenomena, thus facilitating a better interpretation of observational data from detectors like LIGO and Virgo.