GPU accelerators continue to evolve, both in features and capabilities as well as vendor varieties. Simultaneously, supercomputers (which are enormous users of GPUs) also continue to evolve, setting up the very real design challenge of enabling code performance and portability across current and future accelerators.

In this episode of Intel's Code Together podcast, ALCF's Nevin Liber joins Intel's James Reinders and Tony Mongkolsmai to discuss this dilemma, including several possible ways through it, and how researchers are preparing for Argonne's Aurora exascale supercomputer by using C++, Kokkos, and SYCL to enable exascale workloads that are performant now and into the future.