Intel unveiled the first information on general-purpose GPU optimized for HPC (High Performance Compute) and architecture-based systems Xis which will be used in different fields, from artificial intelligence to mobile devices. The code name Old Bridge is a clear tribute to the famous bridge over the Arno river in Florence. The Santa Clara chipmaker has also announced a new scalable and unified programming model.
During the keynote at the 2019 Supercomputing in Denver, Intel explained that the X architectureis it is composed of various micro-architectures, each for each market segment. Ponte Vecchio, based on HPC microarchitecture, uses chiplets made with process technology a 7 nanometers and packaging Foveros. The chiplets are connected to each other via EMIB (Embedded Multi-Die Interconnect Bridge). Communication between GPUs occurs with the CXL (Compute eXpress Link) interface, based on PCI Express 5.0. However, Intel has not provided further details (number of chiplets per GPU, consumption, quantity and type of memory, performance and more).
Ponte Vecchio will be used in the Aurora supercomputer of the Argonne National Laboratory which will be installed in 2021. Aurora will be the first US exascale system based entirely on Intel technology. In addition to the GPU there will in fact be the new 10 nanometer "Sapphire Rapids" Xeon processors and Optane DC memory. Each Aurora node will consist of two Xeon processors and six Ponte Vecchio GPUs. The supercomputer will support over 10 petabytes of memory and over 230 petabytes of storage.
Developers will be able to use the new oneAPI SDK. OneAPI is an open unified programming model for heterogeneous architectures, including CPUs, GPUs, FPGAs and other accelerators. The first beta version supports Xeon processors, Core with integrated graphics and FPGA. The toolkit can be downloaded from place Official.