What Makes HPE Cray’s New EX Supercomputers a Game‑Changer for AI and HPC?

HPE announced an expanded high‑performance computing portfolio that includes several new Cray EX systems and a series of ProLiant Compute servers optimized for artificial‑intelligence workloads.

The flagship EX154n accelerator blade will support up to 224 Nvidia Blackwell GPUs and 8,064 Grace CPU cores per rack, delivering over 10 petaFLOPS FP64 performance for HPC and more than 4.4 exaFLOPS FP4 for sparse AI and machine‑learning tasks.

Each EX154n blade is built around a pair of 2.7 kW Grace‑Blackwell (GB200) super‑chips; every chip integrates two Blackwell GPUs and a 72‑core Arm CPU, interconnected via Nvidia’s NVL4 reference configuration.

Because a single rack can consume more than 300 kW, the system employs a fan‑less, liquid‑cooling solution that extends to all components, including the new Slingshot 400 Ethernet NICs, cables, and switches, which double bandwidth to 400 Gbps compared with the previous 200 Gbps generation.

Both the EX154n and the upcoming CPU‑centric EX4252 Gen 2 compute blade (featuring up to eight 192‑core Turin‑C CPUs, totaling 98,304 cores per rack) are slated for shipment by the end of 2025.

HPE also plans to release an upgraded E2000 storage system that leverages PCIe 5.0 NVMe to more than double I/O performance over prior generations, with availability expected in early 2025.

Complementing the Cray EX line, HPE introduced the ProLiant Compute XD series (XD680, XD685, XD688). The XD688 can be configured with eight Nvidia H200 SXM Tensor‑Core GPUs or Blackwell GPUs, offering up to 1.5 TB of HBM3e memory and employing liquid cooling, with a launch slated for early 2025.

The XD685 variant provides eight AMD Instinct MI325X accelerators and two AMD EPYC CPUs, while the XD680 uses eight Intel Gaudi 3 AI accelerators, both targeting cost‑effective AI training and inference workloads; these models are expected to ship within the next few months.

All new XD servers integrate HPE’s Integrated Lights‑Out (iLO) remote‑management technology for enhanced security and ease of deployment.

HPE’s senior vice‑president for HPC and AI infrastructure, Trish Damkroger, highlighted growing customer interest in autonomous‑AI initiatives that require powerful on‑premise hardware, positioning the new Cray and ProLiant offerings as key enablers for such projects.

Reference: https://www.theregi ster.com/2024/11/13/hpe_cray_ex/ (original source: Semiconductor Industry Observation).