Intel CPUs and GPUs on Structured-Mesh Stencil Workloads with oneAPI
oneAPI and SYCL are quickly evolving to become a key tool for programming a wide breadth of massively parallel architectures. In this talk we present performance results from some of Intel’s latest parallel architectures including CPUs and GPUs on a set of industrially representative and academic structured mesh stencil applications. We benchmark and contrast different formulations of parallel operations on different platforms with both low-order bandwidth-bound kernels as well as high-order more computationally expensive kernels. Finally, we demonstrate scalability by combining MPI and SYCL.
Istvan Reguly
Mark Lubin
Xiao Zhu
