HPE Expands Direct Liquid-Cooled Supercomputing Solutions, Introduces Two AI Systems

Hewlett Packard Enterprise has announced its new high-performance computing and artificial intelligence infrastructure portfolio. It includes HPE Cray Supercomputing EX solutions and two systems optimised for large-language model training, natural language processing and multi-modal model training. 

The portfolio is based on 100% fan-less direct liquid cooling system architecture and offers various supercomputing solutions:

• HPE Cray Supercomputing EX4252 Gen 2 Compute Blade: Capable of delivering up to 98,304 cores in a single cabinet, it features eight 5th Gen AMD EPYC processors. The compute blade offers CPU density, allowing customers to get higher-performing compute within the same space. 

• HPE Cray Supercomputing EX154n Accelerator Blade: To reduce the time for a supercomputing workload to finish, the accelerator blade can accommodate up to 224 Nvidia Blackwell GPUs in a single cabinet. It features the Nvidia GB200 Grace Blackwell NVL4 Superchip, and each blade holds four NVLink-connected Blackwell GPUs unified with two Grace CPUs over NVLink-C2C.

• HPE Slingshot Interconnect 400: The exascale-capable interconnect portfolio offers network interface controllers, cables and switches at 400 gigabit-per-second speeds. It offers features like automated congestion management and adaptive routing for low tail latency.

• HPE Cray Supercomputing Storage Systems E2000: The high-performance storage system more than doubles the input/output performance compared to the previous generation. It is based on the open-source Lustre file system and enables better utilisation of both CPU and GPU-based compute nodes. 

• HPE Cray Supercomputing User Services Software: The new software offering includes features for optimising system efficiency, regulating power consumption and running diverse workloads on supercomputing infrastructure. 

HPE has also rolled out new ProLiant Compute XD servers for streamlining deployment of large, highly performant AI clusters. These include: 

• HPE ProLiant Compute XD680 server: The air-cooled server is designed to address demanding AI training, tuning and inferencing workloads. An HPE-designed chassis houses eight Intel Gaudi 3 AI accelerators in a single compact node. 

• HPE ProLiant Compute XD685 server: The server is powered by eight Nvidia H200 SXM Tensor Core GPUs or Blackwell GPUs in a five rack-unit chassis and leverages liquid cooling for GPUs, CPUs and switches.