HPE Cray Storage is a specialized storage solution engineered for extreme performance in high-performance computing (HPC) and artificial intelligence (AI) workloads. Unlike traditional systems, it handles massive data volumes with fast movement between compute nodes and storage, ideal for AI, climate modeling, and large-scale analytics.
HPE Cray Storage and Parts : Save up to 80%
HPE Cray Storage systems are modular, allowing independent scaling of compute, storage, or cache. With enclosures, compute nodes, and fast interconnects, they offer flexible, high-performance solutions for HPC and AI.
Following HPE’s acquisition of Cray in 2019, this system has evolved into a cornerstone of exascale computing, powering some of the world’s fastest supercomputers, including Frontier, El Capitan, and Aurora.
HPE Cray Storage: Key Differences from Traditional Storage Solutions
HPE Cray Storage stands out from regular storage systems with several advanced features:
HPE Cray Storage Parallel File Systems: It leverages parallel file systems such as Lustre, IBM Spectrum Scale (GPFS), or Cray ClusterStor, allowing multiple nodes to read and write data simultaneously. This parallelism enhances throughput and reduces bottlenecks, critical for HPC and AI tasks.
HPE Cray Storage Data Tiering: Intelligent tiering automatically shifts "hot" (frequently accessed) data to high-speed NVMe SSDs for rapid access, while "cold" (less-used) data is stored on cost-effective SAS HDDs. This optimizes both performance and cost.
HPE Cray Storage Modular Design: Compute and storage nodes can be scaled independently, providing a tailored balance of performance and capacity to meet diverse workload needs.
HPE Cray Storage Designed for Exascale Systems: Cray Storage underpins exascale supercomputers like Frontier (Oak Ridge National Laboratory), El Capitan (Lawrence Livermore National Laboratory), and Aurora (Argonne National Laboratory), showcasing its capability to handle the most advanced computational challenges.
HPE Cray Storage is highly modular, allowing users to scale compute, storage, or cache resources independently based on their workload needs.
Key Components of HPE Cray Storage:
HPE Cray Storage Enclosures that house compute, storage, or cache nodes
HPE Cray Storage Compute Nodes designed for high performance
HPE Cray Storage Interconnect Solutions that enable fast data transfer
HPE Cray Storage Enclosures (Chassis Systems)
Enclosures serve as the foundational chassis for Cray Storage, housing various nodes and providing power, cooling, and connectivity. Here’s a detailed look at the primary enclosures:
HPE Cray ClusterStor E1000 (Storage Enclosure)
Purpose: The primary storage solution for HPC and AI workloads, designed for scalability and high throughput.
Supported Drives: NVMe SSDs (1.92TB to 30.72TB) for high-speed access and SAS HDDs (12TB to 22TB) for capacity.
File Systems: Supports Lustre, IBM Spectrum Scale, and Cray ClusterStor.
Throughput: Delivers up to 85 GB/s read and 65 GB/s write per chassis.
Capacity: Scales from terabytes to exabytes.
Networking: Integrates HPE Slingshot (200 Gb/s Ethernet) and InfiniBand (up to 400 Gb/s).
Cooling: Standard air-cooling with optional liquid cooling for high-density deployments.
HPE Cray Storage Systems C500 (Storage Enclosure)
Purpose: A cost-effective storage solution for entry-level and mid-range HPC/AI clusters, launched in May 2024 to democratize E1000-class technology.
Supported Drives: NVMe SSDs (1.92TB to 30.72TB) for high-speed access and SAS HDDs (12TB to 22TB) for capacity.
File Systems: Supports Lustre, IBM Spectrum Scale, and Cray ClusterStor.
Throughput: Delivers up to 80 GB/s read and 60 GB/s write per 3U chassis.
Capacity: Scales to 2.6 PB (all-flash) or 4 PB (hybrid).
Networking: Integrates HPE Slingshot (200 Gb/s) and InfiniBand (up to 400 Gb/s).
Cooling: Standard air-cooling with optional liquid cooling for dense setups.
HPE Cray Supercomputing Storage Systems E2000 (Storage Enclosure)
Purpose: Next-generation storage for large-scale supercomputers, doubling E1000’s I/O performance; available early 2025.
Supported Drives: NVMe SSDs and SAS HDDs (specific capacities TBD).
File Systems: Optimized for Lustre.
Throughput: Delivers up to 190 GB/s read and 140 GB/s write per chassis.
Capacity: Scales to meet exascale demands.
Networking: Supports HPE Slingshot (200 Gb/s, 400 Gb/s option by Fall 2025) or InfiniBand (400 Gb/s).
Cooling: Primarily liquid-cooled, aligning with exascale systems.
HPE Cray DataWarp (Cache Enclosure)
Purpose: A high-performance caching system for latency-sensitive applications, such as GPU-driven workloads.
Supported Drives: NVMe SSDs only (1.92TB to 30.72TB).
File Systems: Compatible with Lustre, Cray ClusterStor, and IBM Spectrum Scale (GPFS).
Networking: Supports HPE Slingshot (200 Gb/s) or InfiniBand (400 Gb/s).
Cooling: Air-cooled standard, with compatibility for Direct Liquid Cooling (DLC) systems.
HPE Cray EX4000 (Compute Enclosure)
Purpose: Houses high-performance compute nodes for AI and HPC workloads.
Capacity: Supports up to 64 compute blades per cabinet.
Networking: Features HPE Slingshot or InfiniBand.
Cooling: Primarily liquid-cooled, with Direct Liquid Cooling (DLC) options.
HPE Cray Compute Nodes
Compute nodes are specialized servers typically installed within the EX4000 enclosure, designed to deliver exceptional processing power for AI and HPC workloads.
HPE Cray XD665 (Compact Compute Node)
HPE Cray CPUs: 5th Gen Intel Xeon or AMD EPYC (up to 2 CPUs per node).
HPE Cray GPUs: Up to 4 NVIDIA H100 Tensor Core GPUs (80 GB HBM3 each).
HPE Cray Memory: Up to 1.5 TB DDR5-6400 MT/s.
HPE Cray Networking: Dual-port HPE Slingshot or InfiniBand HDR (400 Gb/s).
HPE Cray Cooling: Air-cooled or Direct Liquid Cooling (DLC).
HPE Cray XD670 (High-Performance Compute Node)
HPE Cray CPUs: 5th Gen Intel Xeon (up to 2 CPUs per node).
HPE Cray GPUs: Up to 8 NVIDIA H100 (80 GB HBM3) or H200 (141 GB HBM3e) GPUs.
HPE Cray Memory: Up to 1.5 TB DDR5-5600 MT/s.
HPE Cray Cooling: Direct Liquid Cooling (DLC).
HPE Cray XD675 (AI-Optimized Compute Node)
HPE Cray CPUs: 5th Gen AMD EPYC (up to 2 CPUs per node).
HPE Cray GPUs: Up to 8 AMD Instinct MI300X GPUs (192 GB HBM3 each).
HPE Cray Memory: Up to 1.5 TB DDR5-6400 MT/s.
HPE Cray Cooling: Direct Liquid Cooling (DLC).
HPE Cray EX154n (Extreme GPU Compute Node)
HPE Cray CPUs/GPUs: NVIDIA GB200 Grace Blackwell Superchip (4 GPUs + 2 CPUs per Superchip).
HPE Cray GPU Capacity: Up to 224 NVIDIA Blackwell GPUs per cabinet.
HPE Cray Memory: Scalable DDR5 capacity integrated with the NVIDIA GB200 Superchip.
HPE Cray Cooling: 100% Direct Liquid Cooling (DLC).
HPE Cray Storage and Cache Nodes
Storage and cache nodes are designed to handle data movement and management efficiently. Typically both are integrated in the HPE Cray ClusterStor E1000 storage enclosure.
Storage Nodes
HPE Cray Drives: NVMe SSDs (1.92TB to 30.72TB) and SAS HDDs (12TB to 22TB).
HPE Cray File Systems: Lustre, IBM Spectrum Scale, and Cray ClusterStor.
HPE Cray Throughput: Up to 85 GB/s read and 65 GB/s write per 3U chassis.
Cache Nodes
HPE Cray Drives: NVMe SSDs only (1.92TB to 30.72TB).
HPE Cray Networking: HPE Slingshot or InfiniBand.
HPE Cray File Systems: Lustre, GPFS, or Cray ClusterStor integration.
HPE Cray Storage Management Nodes
Management nodes oversee system administration, monitoring, and maintenance. It's typically integrated in the HPE Cray EX4000 compute enclosure.
HPE Cray Software: HPE Cray System Management (CSM), a centralized interface for hardware status monitoring, firmware updates, and data movement.
HPE Cray Networking: Integrated with HPE Slingshot or InfiniBand.
HPE Cray Storage Interconnects
High-speed interconnects ensure fast communication between compute, storage, and management nodes. HPE Cray interconnects — HPE Slingshot and InfiniBand HDR — are built directly into compute nodes, storage nodes, cache nodes, and management nodes, providing fast data movement across the entire system.
HPE Cray Slingshot: A 200 Gb/s Ethernet fabric designed for HPC, supporting Dragonfly Topology for large-scale clusters.
HPE Cray InfiniBand HDR: Up to 400 Gb/s, providing low-latency, high-bandwidth connectivity.
HPE Cray Topology: Dragonfly Topology ensures efficient data movement in expansive systems.
HPE Cray Storage and Parts : Save up to 80%
Comments