Kestrel and Adastra will both be based on the Cray EX supercomputing platform.
Hewlett Packard Enterprise (HPE) was recently awarded two major contracts to deliver cutting-edge supercomputers to high-profile contacts: the U.S. Department of Energy and France’s National Computing Center for Higher Education (CINES).
The Department of Energy’s National Renewable Energy Laboratory (NREL) will use HPE’s Kestrel supercomputer to turbocharge renewable power research. Kestrel will support the laboratory’s research and development into technologies that will harness energy from geothermal, water, wind, solar, as well as fuels.
The Kestrel will be built using the powerful HPE Cray EX supercomputer. The Cray is a liquid-cooled, blade-based high-density clustered computer system. Its basic component is a sealed liquid-cooled cabinet that uses closed-loop cooling technology to avoid exhausting hot air in a data center. Liquid cold plates are attached directly to high power devices such as processors, GPUs and switches to remove the heat more efficiently via an auxiliary cooling distribution unit.
Kestrel will use Intel Xeon Scalable (“Sapphire Rapids”) server processors—which integrate High Bandwidth Memory and will be the chipmaker’s highest performance data center processor yet (the chip is expected to be released in 2022). Kestrel will also use NVIDIA A100NEXT Tensor Core GPUs to boost artificial intelligence (AI) functionality, as well as HPE Slingshot—an Ethernet fabric specifically designed for next-generation supercomputers to better handle higher speed and congestion control for the data-intensive and AI workloads the supercomputer is expected to handle.
Kestrel will deliver more than 75 petabytes of parallel file system storage using HPE’s Cray ClusterStor E1000 storage system, which will further enable expanded storage and intelligent tiering capabilities.
Kestrel will provide NREL with a high-performance computing (HPC) solution that will deploy advanced modeling, simulation, AI and analytics capabilities to generate more efficient insights across the Department of Energy’s research initiatives—which include research in computation materials, continuum mechanics and engaging in the large-scale simulation, and planning for the energy systems of the future.
The supercomputer will harness the power of rapidly advancing AI and machine learning technologies as well as synthesize virtual simulations with physics-based sensor data sources using a heterogeneous architecture that includes CPU-only as well as GPU-accelerated nodes to process that information. Kestrel’s operators are hoping that the computer will help accelerate innovations and provide solutions to the challenges of moving to renewable and sustainable energy sources.
NREL will also use the supercomputer itself as an example of energy efficiency. Kestrel will be an integral part of NREL’s showcase Energy Systems Integration Facility data center in Golden, Colo., which is a leader in data center efficiency. In fact, NREL has achieved a yearly average power usage effectiveness (PUE) rating of 1.036. PUE is the ratio of the total amount of power used by the data center facility to the power used by the actual computing equipment. So in the case of NREL, almost all of the incoming power is dedicated to the computer. To put this in context, a typical data center has an average PUE of around 1.8—and that number can fluctuate across a wide range.
“HPE has a long-standing collaboration with the National Renewable Energy Laboratory where we have developed joint high-performance computing and AI solutions to innovate new approaches that reduce energy consumption and lower operating costs,” said Bill Mannel, HPE’s vice president and general manager, HPC. “We look forward to continuing our relationship with NREL and are honored to have been selected to deliver an advanced supercomputer with Kestrel that will significantly augment the laboratory’s efforts in making breakthrough discoveries of new, affordable energy sources to prepare for a sustainable future.”
Installation will begin in 2022. When it comes online in 2023, Kestrel will have about 44 petaflops of peak performance—that’s more than five times greater than the performance of NREL’s existing Eagle supercomputer.
But Kestrel isn’t the only new supercomputer in the works for HPE. The company is also building what will be one of Europe’s most powerful supercomputers at France’s GENCI, the country’s national agency for High Performance Computing and CINES (National Computing Center for Higher Education)—one of only three HPC centers in France.
This machine will be known as “Adastra” and will also be based on the Cray EX system, but instead of being powered by Intel, the supercomputer will use third generation AMD EPYC server processors and an accelerated compute with the newly launched AMD Instinct MI200 accelerators. These technologies will take on data and image-intensive workloads to provide scientists and researchers with faster and more accurate insights.
Adastra will be used for targeted research areas and will support the development of renewable energy to bolster France’s energy security and reduce the country’s need to import fuel, as well as conserve natural resources. The computer will also be deployed to discover new, highly effective materials for creating next-generation batteries. And in the health care field, the computer will be used to advance medical research to aid drug design and medical treatments.
“We are honored to have been selected by GENCI, which empowers some of the world’s leading research centers, to deliver advanced high-performance computing and AI solutions using exascale-era technologies in support of propelling France’s R&D efforts,” said Justin Hotard, HPE’s senior vice president and general manager, HPC and AI.
Loaded up with cutting-edge HPC and AI technologies, Adastra will be able to deliver more than 70 petaflops of performance—that’s a rate more than 20 times faster than the organization’s existing supercomputer—for processing large-scale complex scientific data.
The Cray EX-powered system will deliver end-to-end exascale era technologies across compute, software, storage and networking. This includes powerful and accelerated compute to provide sharper modeling, simulations and data and image-intensive workflows in two partitions. The first will use optimized third generation AMD EPYC processors combined with AMD Instinct MI250X accelerators, each with 128 GB HBM2e, for a total of 512 GB GPU memory. The other partition will use “Genoa” AMD EPYC processors with 768 GB of DDR5 memory. Like Kestrel, Adastra will also use the HPE Slingshot Ethernet fabric and ClusterStor E1000 storage system.
Adastra will feature the HPE Performance Cluster Management solution for managing the system and enabling fine-grained centralized monitoring and management for optimal performance, as well as a fully integrated software suite to optimize HPC and AI applications using HPE’s Cray Programming Environment.
And also like its Kestrel cousin, Adastra is being designed to be as energy efficient as possible. In all, 97 percent of the heat generated by the supercomputer will be liquid cooled. The solution has the potential to achieve a PUE of 1.10 with a maximum sustained consumption of only 1.59 megawatts.
“This new Adastra supercomputer is double challenging for CINES: massively switch users to boosting GPU while overcoming the electrical consumption. The 21-times jump in computing power is dizzying compared to the current machine, while power consumption only grows by 1.5 times for green HPC,” said Boris Dintrans, director of CINES. “The GPU support provided by AMD, using the HPE Cray EX system, allowing Adastra to ramp up, is the key element that will ensure the success takeoff of researchers to the stars!”
Adastra will be based in CINES, in Montpellier, France and the system is anticipated to be online in 2022.
HPE is already an in-demand powerhouse in cutting-edge computing. With offerings like Kestrel and Adastra, the company continues to push the envelope on what a supercomputer can achieve. And by factoring energy use considerations into its projects, HPE is opening new avenues for technological innovation in energy consumption as well.
Read more about how supercomputing can transform your work at High-Performance Computing 101: HPC and How Engineers Can Use It.