At SC20, Intel announced that it is making its Xe-HP high-performance discrete GPUs available for early access developers. Specifically, the new chips have been deployed at the Argonne National Laboratory and will serve as the transition development vehicle for the future Aurora (2022) supercomputer that will support the delayed Intel Xe HPC GPUs (“Ponte Vecchio”) intended for computing the backbone Systems.
The Xe-HP-based development platforms are used by the Argonne Leadership Computing Facility (ALCF) as part of the Aurora Early Science Program and the Exascale Computing Project to ensure that applications, libraries and infrastructures are exascale-enabled.
“Our collaboration with Intel on cross-architecture code development benefits many of our development teams,” said Susan Coghlan, project leader for Aurora at ALCF. “This co-design approach has resulted in the software stack rapidly maturing to production quality for execution on Aurora.”
Aurora is built through a partnership between Intel, Hewlett Packard Enterprise and Argonne National Laboratory. The Exascale-class system implements HPE’s Cray EX supercomputer architecture with Slingshot networks, a future generation of persistent Intel Optane memory and Intel oneAPI software.
Aurora’s node design includes two future-generation 10nm ++ “Sapphire Rapids” CPUs (with improved SuperFin technology) and six “Ponte Vecchio” Xe HPC data center GPUs.
The schedule and the exact design of the Xe HPC GPUs are still unclear. However, after Intel announced a one year delay due to a failure mode in the 7nm process, Intel has now committed to delivering Aurora in 2022. A previous incarnation of Aurora was originally slated to be deployed in 2018, but that project was recast in 2017 as the country’s first exascale machine with a target for 2021.
At the Intel Architecture Day in August, the company announced that the Xe-HPC-GPU will be manufactured with its 10-nm SuperFin for the basic tile and 10-nm-Enhanced SuperFin for the Rambo cache tile. However, the company has not disclosed the process node for the compute tile, which is either using an Intel Next Gen process or going through an external fab.
The Xe-HP GPU was implemented in 1-, 2- and 4-tile designs made with 10 nm Enhanced Superfin. The 4-tile version offers over 40 teraflops of FP32 performance (according to Intel). The 2-tile version that Intel provided for Argonne offers around half of this performance (~ 21 teraflops of FP32).
Xe HP GPU with two tiles
In an Intel keynote presentation at the SC20, Trish Damkroger, general manager of Intel’s HPC Group, said the collaboration between Intel and Argonne will focus on the development of next generation semiconductor technologies, manufacturing processes, advanced system design and software enablement, including future silicon, will focus on development, future architectures for high performance computing and AI as well as the activation of software ecosystems for exascale computing.
“Aurora will provide scientists and researchers with an unprecedented suite of tools and applications that will enable breakthrough advances in a variety of areas that will benefit us all, including medicine, weather modeling, climate change and materials science,” he said.
The Xe-HP development platforms support the co-design, testing and validation of multiple exascale applications, including:
The EXAALT project, which enables molecular dynamics in the exascale for fusion and fission energy problems.
The QMCPACK project developing Exascale Quantum Monte Carlo algorithms to improve predictions about complex materials.
The GAMESS project, which develops ab initio fragmentation methods to address challenges in computational chemistry such as heterogeneous catalysis problems more efficiently.
The ExaSMR project, which develops high-fidelity modeling functions at exascale for complex physical phenomena that occur in the operation of nuclear reactors in order to ultimately improve their design.
The HACC project, which develops on exascal extreme-scale cosmological simulations that allow scientists to simultaneously analyze observational data from state-of-the-art telescopes to test various theories.
Related reporting: Intel introduces oneAPI Gold and provides further details on the GPU roadmap