This blog post discusses a new paper on executing Fortran code on GPUs without needing to use external APIs or directives. It highlights the use of Fortran’s "do concurrent" feature, which facilitates running code on NVIDIA, AMD, and Intel GPUs, with mentions of efficiency improvements through OpenMP directives for Intel GPUs and requirements for AMD GPUs. It emphasizes a Fortran-only approach that is effective across multiple multi-GPU nodes.