Based on challenging requirements like these, Cray developed the XC series supercomputers to easily support differing processor technologies in the same architecture. By leveraging the advantages of continuously changing multicore and many-core processing devices, Cray system owners can target the most appropriate technologies to get the best possible performance out of differing applications.
The Swiss National Supercomputing Centre (CSCS) is completing a massive upgrade of its Cray XC series supercomputer Piz Daint, marking a key example of “adaptive” supercomputing and introducing a significant performance leap with the integration of new, leading-edge accelerators.
Cray’s new XC50 product line leverages the productivity of a rich software development environment and the power of the legendary Cray Aries interconnect-network technology while unleashing the next wave of best-in-class processors and accelerators to help expand the bounds of computation. The XC series (and our newest entry, the XC50) expands our support for the latest CPU and CPU capabilities with the addition of NVIDIA’s Tesla P100 GPUs, previously code-named “Pascal.”
The XC50 is the fastest Cray supercomputer to date, with one PF/system cabinet peak performance.
The new Cray XC series P100 compute blade architecture supports four compute nodes, each coupling a host processor with a NVIDIA Tesla P100 GPU accelerator. To better support a diverse range of high-performance applications, the P100 offers flexible user operation modes for double-precision, single-precision, or half-precision applications and integrates high-bandwidth memory on board the device. These new features will be particularly advantageous for applications with intense memory bandwidth performance demands as well as deep learning applications and traditional HPC apps. In fact, the Cray XC series implementation of the P100 GPU enables up to a remarkable one PF of peak performance per XC50 cabinet.
Cray dominates the total GPU performance of the world's top machines, and the company’s broad software development environment has a long history of providing multiple methods for programming CPU-GPUs systems, including CUDA and directives-based models. The Cray software environment also includes optimized libraries, productivity tools for debugging, performance analysis, and porting, as well as access to other tools and libraries from the ecosystem of partners and open organization.
Real-World Applications Drive Discovery
Cray works closely with amazing research centers like CSCS to help them drive modern discovery via real-world applications (see their Cray upgrade story from earlier this year), with a focus on optimizing “time-to-solution” and “energy-to-solution.” To successfully push the frontiers of science and engineering, Cray systems are designed to deliver sustained performance and scalability, leveraging the adaptive nature of integrated multicore and many-core technologies. CSCS helps guide Cray on future requirements so that we can both satisfy the users of Piz Daint across their numerous application areas – chemistry, earth and environmental science, engineering, life sciences, materials science, mechanics, physics, and even seismology. The CSCS/Cray collaboration on the Piz Daint Pascal upgrade is a success model for upgradability by design.
Industry benchmarks can provide a starting point for general system comparisons, and many top systems are well documented publicly, with the Cray XC series Piz Daint system entering the top 10 of the Top 500 list in November 2013 (at number six) and holding the place of the top system in Europe since then. Keep an eye on CSCS as it updates its European leading performance benchmarks for the TOP500 and Green500 for their upgraded Piz Daint system this month.
Grander, more complex and even previously unaddressable challenges are always around the next corner. Cray is excited to introduce our highest-performing supercomputer architecture to date and eagerly watches to see what CSCS and all its extended users can do with this giant leap in Cray’s adaptive supercomputing platform.