Cray

HPE Cray Supercomputing

HPE Cray supercomputing stands out by unifying high-performance computing (HPC) and artificial intelligence into a single, cohesive architecture. Unlike traditional systems, it utilizes the HPE Slingshot interconnect, a purpose-built Ethernet fabric that provides high-speed, low-latency communication while using advanced congestion management to prevent data bottlenecks. This allows the systems to handle massive, data-intensive AI workloads and complex scientific simulations simultaneously without compromising performance.

Engineered for extreme scale, these systems feature industry-leading compute density and Direct Liquid Cooling (DLC), which allow for a smaller physical footprint and significantly lower energy consumption compared to air-cooled alternatives. By integrating a specialized software stack and high-performance storage, HPE Cray eliminates the hardware trade-offs typically found in large-scale computing, making it the primary architecture used for the world’s most powerful exascale supercomputers.

HPE Cray Supercomputing offers significant sustainability benefits by combining high-density architecture with advanced thermal management. A key feature is the use of direct liquid cooling (DLC), which is up to 100 times more efficient at heat transfer than air. By eliminating energy-intensive server fans, DLC can reduce total energy consumption by approximately 10% and cooling-specific power by up to 37%. Furthermore, the warm water generated by these systems can be repurposed for waste heat recovery to heat surrounding buildings or greenhouses, potentially offsetting the facility’s overall carbon footprint.

Beyond operational efficiency, HPE Cray systems help reduce the environmental impact of physical infrastructure. The high-density design can halve the required data center floor space, lowering the embodied carbon associated with building construction. Organizations can also leverage the HPE Sustainability Insight Center for real-time monitoring of energy use and carbon emissions, while HPE Asset Upcycling Services ensure that retired hardware is refurbished or recycled, supporting a circular economy and keeping materials out of landfills.

To get started with an HPE Cray supercomputer deployment, organizations must first complete site preparation, ensuring the facility can support the required power loads, cooling requirements, and floor-weight capacities. This process is guided by the HPE Cray Supercomputer Startup Service, which assigns a dedicated HPC Installation Project Manager to oversee the transition from factory integration to onsite setup. This service delivery manager coordinates the physical installation, including uncrating, rack positioning, and power-on verification, so all environmental prerequisites are met before the hardware arrives.

Once the physical infrastructure is in place, the deployment shifts to initializing the HPE Cray System Management (CSM) framework and the HPE Cray Programming Environment (CPE). Administrators use tools like the Install and Upgrade Framework (IUF) to configure the operating system, set up optimized compilers and libraries, and integrate workload managers such as Slurm or PBS Professional. This structured approach provides the high-speed interconnects and management services are fully tuned and ready for production-level HPC and AI workloads.

Since acquiring Cray Inc. in 2019, HPE has preserved the legacy of Seymour Cray by maintaining HPE Cray as its premier brand for supercomputing. By retaining the name for its most advanced systems, HPE honors the brand’s history of innovation while integrating Cray's specialized engineering talent and proprietary technologies—such as the Slingshot interconnect and high-density liquid cooling—into its core product portfolio.

This stewardship has enabled the successful launch of the world’s first exascale supercomputers, including El Capitan and Frontier systems. By evolving Cray's architectural principles to meet modern AI and data science demands, HPE keeps the name synonymous with cutting edge innovation for scientific discovery and national security technology.

The primary advantage of HPE Cray is its purpose-built architecture, specifically designed to eliminate the bottlenecks found in standard clusters. While many competitors rely on general-purpose networking, HPE Cray features the Slingshot interconnect, which provides high-speed, congestion-controlled data movement. This allows thousands of processors to work together as a single, unified machine without slowing down, making it the superior choice for massive AI model training and complex scientific simulations that outgrow traditional hardware.

Furthermore, HPE Cray systems lead the industry in energy efficiency and density through advanced direct liquid cooling. By cooling every component with liquid rather than fans, these systems reduce power consumption and floor space requirements while supporting the highest-performance CPUs and GPUs. This exascale-ready engineering is the same technology powering the world's fastest supercomputers, offering a proven, reliable environment for organizations tackling the most data-intensive challenges.

HPE Cray services provide a comprehensive, turnkey experience that removes the complexity of managing exascale-class hardware. Unlike standard IT support, these services are delivered by specialized experts who understand the nuances of massive parallel processing and high-density liquid cooling. From initial installation to 24/7 operational monitoring, the support team ensures that these massive systems remain optimized for the highest possible uptime, allowing organizations to tackle grand challenge problems without needing to manage every intricate hardware detail themselves.

Beyond physical maintenance, the service experience includes a specialized software stack designed to maximize researcher productivity. Tools like the HPE Cray Programming Environment and advanced management software automate system provisioning and optimize data flow, so system noise doesn't interfere with complex calculations. This proactive approach—combining AI-driven monitoring with deep domain expertise—keeps  the architecture remains peak-efficient, allowing users to focus on scientific discovery and model training rather than troubleshooting.