HPE’s new AI servers tap Nvidia, AMD, Intel GPUs for supercomputer flexibility

HPE’s new AI servers tap Nvidia, AMD, Intel GPUs for supercomputer flexibility

Close up of Hewlett Packard Enterprise logo and sign on the building in San Jose, Ca.
JHVEPhoto/Adobe Stock

Hewlett Packard Enterprise (HPE) has unveiled new supercomputing solutions designed to power intense workloads like AI language and multi-modal model training featuring innovative cooling features.

Designed to help customers accelerate research and innovation efforts, HPE launched new supercomputing servers powered by GPUs from the likes of Nvidia and AMD, as well as software built to make it easier for users running intensive workloads.

The portfolio is based on HPE’s recently unveiled fanless direct liquid cooling system architecture, which the company claims provides a 37% reduction in cooling power per server blade compared to hybrid direct liquid cooling.

Subscribe today for free

At the top end of the portfolio are two new Cray EX solutions, the EX4252 and EX154n, each sporting accelerators from rival chip designers.

EX4252 is a one-rack unit that features eight of AMD’s new 5th gen EPYC processors that is capable of delivering up to 98,304 cores in a single cabinet.

The EX154n, meanwhile, is designed to hold up to 224 Nvidia Blackwell GPUs in a single cabinet. Featuring Nvidia’s GB200 Grace Blackwell Superchip, HPE said it’s built to “drastically reduce the time it takes to complete a supercomputing workload.”

The EX4252 will be available in Spring 2025, while customers will have to wait until the end of 2025 to get their hands on the EX154n as Nvidia works to ramp up Blackwell production.

Also unveiled by HPE are the Slingshot interconnect 400, designed to deliver twice the line speed over the previous generation, while also offering features like automated congestion management and adaptive routing for ultra-low tail latency — enabling users to run large workloads with significantly less network infrastructure.

Another expansion of the Cray Supercomputing portfolio includes the addition of the Storage Systems E2000, a high-performance storage system designed specifically for large-scale supercomputers. HPE claims it more than doubles E2000 the input/output (I/O) performance compared to storage solutions, enabling better utilisation of both CPU and GPU-based compute nodes by reducing idle time during I/O operations.

The E200 will be made available in early 2025, with the Slingshot interconnect 400 to be released later in the fall.

HPE also introduced a new software designed to aid user experiences when running compute-intensive workloads.

The HPE Cray Supercomputing User Services Software aims to help customers optimise system efficiency, regulate power consumption, and flexibly run diverse workloads on its portfolio of supercomputing infrastructure solutions.

Supporting the supercomputing solutions are two new HPE servers specifically designed to power AI workloads.

The ProLiant Compute XD680 server can power AI training, tuning and inferencing workloads, with its chassis housing eight Intel Gaudi 3 AI accelerator in just one node.

The ProLiant Compute XD685 server, meanwhile is designed to accelerate training for large, complex AI models. Built featuring either Nvidia H200 or Blackwell GPUs in a five rack-unit chassis, the XD685 is suitable for users “prioritising performance, competitive advantage and energy efficiency,” an HPE announcement suggests.

The XD680 will be available in December 2024, while the H200-powered XD685 will launch in early 2025, with Blackwell versions being made available at a later date.

A version of the XD685 server featuring AMD Instinct MI325Xs, however, will be available, in the first quarter of 2025.

“Our customers turn to us to fast-track their AI system deployment to realise value faster and more efficiently by leveraging our world-leading HPC solutions and decades of experience in delivering, deploying and servicing fully integrated systems,” said Trish Damkroger, SVP and general manager for HPC and AI infrastructure solutions at HPE.

RELATED STORIES

HPE launches AI-driven network detection and response capabilities

HPE unveils end-to-end AI-native portfolio for GenAI

HPE to acquire Juniper Networks for $14bn

Gift this article