Intel's first Xe-HPC graphics processor for supercomputing applications codenamed Ponte Vecchio has not taped out yet, but developers of programs for the upcoming Xe-HPC-based Aurora supercomputer need to start writing and testing their code now in a bid to use it sometime in 2022. To enable software development for the new machine, Intel is getting ready to provide select Aurora developers DevCloud access to its Xe-HP-based systems.
Intel is preparing to add support for Xe-HP to its Linux graphics drivers, which is required both for Aurora developers that will design programs for the ExaScale-class supercomputers as well as other parties that plan to use Xe-HP hardware in their datacenters. Interestingly, Intel's Linux drivers refer to Xe-LP discrete and integrated GPUs as 'Gen 12,' whereas, Xe-HP is named 'Gen 12.5,' reports Phoronix.
From a microarchitectural point of view, Intel's Xe-HPC is a superset of Intel's Xe-HP microarchitecture that supports a host of new instructions (and possibly data formats) tailored for supercomputing. Also, Intel's Ponte Vecchio has numerous performance-boosting features, such as Rambo cache and high-speed I/O. All of Intel's Xe microarchitectures (just like CPUs, FPGAs, and ASIC) can be programmed using Intel's oneAPI toolkits. To that end, Xe-HPC can execute code written for Xe-HP, so it's reasonable to the latter as a development vehicle to build and validate software for the former.
It's no secret that Argonne National Laboratory's Aurora supercomputer will be delayed by at least several months because of Intel's setback with its 7 nm fabrication process. Intel also recently said that while the development of its first Xe-HPC GPU had been completed, it still had not received the multi-chiplet processor back from the fab. Meanwhile, it is time for software developers to begin writing programs both for Xe-HP and Xe-HPC hardware, so Intel is providing DevCloud access to Xe-HP.
"Our close collaboration with Argonne is enabling us to make tremendous progress on Aurora, as we seek to bring exascale leadership to the United States," said Trish Damkroger, Intel vice president and general manager of High-Performance Computing. "Providing developers early access to hardware and software environments will help us jumpstart the path toward exascale so that researchers can quickly start taking advantage of the system’s massive computational resources."
Argonne National Laboratory will use Intel's Xe-HP hardware to test and validate multiple programs that will be used on the upcoming Aurora supercomputer. The list includes apps that work on such complex problems as molecular dynamics for fusion and fission energy problems, extreme-scale cosmological simulations, computational chemistry for heterogeneous catalysis problems, and predictions concerning complex materials using Quantum Monte Carlo algorithms at exascale.