3D modelling of the hydrodynamics of stars requires extremely large computational resources. 3D hydrodynamics codes, such as PROMPI, are parrallelised codes, which can run efficiently on tens of thousands of cores using the common MPI system. However, it still take months on thousands of cores to simulate a few convective turnover times for the very high resolutions required to capture the internal dynamics of stellar evolution. These large HPC systems therefore require large amounts of electricity. GPUs provide an energy-efficient alternative to MPI. They provide parallel computation particularly focused on matrices and vectors due to their use in graphics. To use this method requires a different programming paradigm to MPI, although both must be used to solve large problems.
We will present our initial investigations into the potential of using GPUs in the PROMPI code. As a test-bed, the Helmholtz equation of state module in PROMPI is being developed to run on multi-GPUs through the use of CUDA. Although still in the early stages of development, the code is providing machine accurate solutions and speed up in the computational time in this first test. This test shows great promise for the potential of implementing GPU parallelism into the full PROMPI code in the future.