HPE Cray EX235a

HPE Cray EX235a is a liquid-cooled supercomputer node architecture developed by Hewlett Packard Enterprise (HPE) as part of its Cray EX supercomputing platform. It integrates AMD's third-generation EPYC "Milan" central processing unit (CPU) with four AMD Instinct GPU accelerators, designed for high-performance computing (HPC) and artificial intelligence (AI) workloads. The system exemplifies the convergence of traditional computational science with modern machine learning demands within the exascale computing era.

Overview

The HPE Cray EX235a represents a key building block in the Cray EX supercomputing architecture, which itself is the foundation for several of the world's most powerful systems, including the Frontier system at Oak Ridge National Laboratory. It is engineered as an accelerated compute node, combining high-core-count x86-64 server processors with dense GPU acceleration in a tightly integrated, direct liquid-cooled chassis. This design philosophy continues the legacy of Cray Inc. in building systems for leading national laboratories and research institutions, such as those managed by the United States Department of Energy and the European Centre for Medium-Range Weather Forecasts.

Architecture

The node's compute architecture is centered on a single AMD EPYC 7003 series CPU, which utilizes the Zen 3 microarchitecture and is connected via high-bandwidth Infinity Fabric links to four AMD Instinct MI250X or AMD Instinct MI250 accelerator cards. These accelerators are multi-die GPUs featuring Matrix Cores optimized for FP64 and FP32 floating-point arithmetic, as well as specialized AI matrix math operations. The node leverages the Cray Slingshot interconnect, a high-performance Ethernet-based fabric, for low-latency communication across a supercomputer system. All major components, including the CPU, GPUs, and memory, are cooled by a warm-water, direct-to-chip liquid cooling system, which improves thermal efficiency and power density compared to traditional air cooling.

Performance

In configured systems, the HPE Cray EX235a delivers exceptional performance for both double-precision (FP64) scientific simulation and mixed-precision AI training. A single AMD Instinct MI250X accelerator can achieve over 47 teraFLOPS of peak FP64 performance. When deployed at scale, such as in the LUMI system hosted by the CSC – IT Center for Science in Finland, nodes based on this architecture contribute to aggregate performance measured in hundreds of petaFLOPS. The architecture's balance of CPU and GPU power, coupled with the high-throughput Cray Slingshot network, enables strong scaling for applications in fields like computational fluid dynamics and climate modeling.

Software and Programming Environment

The system is supported by the unified HPE Cray Programming Environment, which includes optimized compilers such as the AMD Optimizing C/C++ Compiler (AOCC), libraries like the ROCm open software platform, and performance analysis tools like Cray Performance Measurement and Analysis Tools. This environment supports standard parallel computing models including OpenMP, MPI, and OpenACC, as well as AI frameworks like PyTorch and TensorFlow via ROCm. The Cray Linux Environment, based on SUSE Linux Enterprise Server, provides the core operating system and workload management, often integrated with schedulers like Slurm Workload Manager.

Applications and Use Cases

Primary applications for the HPE Cray EX235a are found in large-scale scientific research and national security. It is deployed for grand challenge problems such as molecular dynamics simulations for drug discovery, cosmological simulations of the universe, and developing advanced weather forecasting models at institutions like the Met Office. The integrated GPU acceleration also makes it a premier platform for training large language models and conducting AI for science research, exemplified by projects at the Argonne National Laboratory and the Lawrence Livermore National Laboratory.

Market Context and Competitors

The HPE Cray EX235a competes in the high-end HPC and AI market segment against other accelerated supercomputing architectures. Key competitors include nodes featuring Nvidia A100 or H100 GPUs paired with Intel Xeon or AMD EPYC CPUs, such as those found in systems from Dell, Lenovo, and Atos. It also contends with Arm-based solutions like the Fujitsu A64FX processor used in the Fugaku system. The Cray EX platform's differentiation lies in its full-stack, liquid-cooled design, the high-bandwidth Cray Slingshot interconnect, and its deep integration heritage from Cray Inc., targeting flagship installations for government and academic research consortia.

Category:Supercomputers Category:Hewlett Packard Enterprise Category:Computer-related introductions in the 2020s