HPE Cray EX

HPE Cray EX

The HPE Cray EX supercomputer is a water-cooled, clustered solution built on high-density blade servers designed to deliver the very highest performance, scalability and compute density. The core building block of the HPE Cray EX supercomputer is either a water-cooled cabinet or a water-cooled rack. Both configurations, using directly installed cold platters, provide efficient heat removal from the most power-hungry parts such as processors, graphics accelerators and switches, all thanks to a cooling distribution unit(CDU).

8

Servers on the system

1.6 MW

Cooling capacity

15 kW

On the rack

64

Blade servers

8

CPU on the blade

384 TB

RAM for cabinet

Hardware Software

Top performance and scalability

HPE Cray EX systems come with revolutionary computing capabilities thanks to their implementation of the latest 4th generation AMD EPYC and 4th generation Intel Xeon Scalable processors. With an architecture that builds on high-density (Node with CPU blade servers) and support for various graphics and compute accelerators such as NVIDIA Tesla and AMD Instinct, these servers can handle complex simulations, data-intensive applications and demanding computations with complete ease.

6 TB DDR5

One node can be equipped with up to 6 TB of DDR5 Smart Memory with ECC support

High-speed NVMe

PCIe Gen 4 servers support high-speed NVMe SSDs

Customizability

The entire system can be tailored to work in any environment

Efficient water cooling

Developed and optimized for maximum efficiency, HPE Cray EX systems have advanced water cooling systems in place to maintain optimal operating temperatures and reduce power consumption. A closed cooling loop ensures minimal impact on the data center environment, which does not need any special modifications.

HPE Cray EX Watercooled rack servers with HPE CDU

Memory and storage

HPE Cray EX systems are designed with versatile memory and storage configurations to meet the needs of high-performance computing tasks. These features ensure that applications can scale efficiently without bottlenecks. The servers can be equipped with up to 64 GB of 4800 MT/s DDR5 DIMMs for AMD EPYC processors, delivering up to 6 TB of capacity per node. At the same time, the servers can fully utilize HBM3 memory on attached GPUs. The servers can be equipped with up to four M.2 NVMe SSDs, and seamlessly connect to high-capacity external storage.

Innovative network infrastructure with HPE Slingshot

HPE Cray EX systems are interconnected using HPE Slingshot interconnect, a high-speed fabric network designed for the performance found in exascale compute clusters. The Slingshot network guarantees low latency and the highest possible throughput, making it ideal for high-demand, diverse work environments. The Slingshot network consists of 64-port HPE Slingshot switch blades with 200Gbps per port, which together form a Dragonfly topology that is highly scalable and manageable through management tools.

Comprehensive software ecosystem

HPE Cray EX systems come with an integrated software package to maximize performance and make management and use as simple as possible. The tools are designed to accommodate everyone from system administrators to developers, so overall system usability is increased and gives room for better optimization. The software suite consists of HPE Cray System Managers for system monitoring and provisioning, as well as HPE Cray Programming Environmen for application development and debugging, while having support for a wide range of virtualization and containerization applications such as Singularity and Docker.

Programming Env Cray Management

Video: HPE Cray XD675

In this video from HPE, we summarize the advantages of the similar Cray XD675 system. This system has been built in partnership with AMD on 4th generation AMD EPYC processors and AMD Instinct MI300 graphics accelerators, resulting in a high performance and highly optimized HPC solution for demanding applications.

Whitepaper