Meet the New HPE ProLiant Compute XD Family Powering Enterprise AI Workloads

[Dima_Tsaripa]

The printing press, the steam engine, the internet, and now artificial intelligence (AI): experts agree that AI is currently transforming the business world as profoundly as – some even argue: more profoundly than – the revolutions caused by those previous cornerstone innovations. Just like the internet needed powerful routers to interconnect the world, AI needs powerful servers for model training and AI usage (inference). HPE now expands its comprehensive AI portfolio with a new family of servers specifically designed for demanding enterprise AI workloads.

AI paves the way for a new generation of innovations while improving numerous processes across industries. With the help of AI, physicians and researchers can detect tumors more reliably and in earlier stages, while robotic prosthetics give patients new levels of mobility. The pharmaceutical idustry can automate molecular research, considerably accelerating drug design. Financial institutions use AI to identify fraudulent transactions and to streamline risk assessment throughout their portfolios. Telco network operators are alerted of potential performance degradation in real-time. And companies from any verticals can streamline call centers and customer services desks by having AI answer up to 50% of incoming questions and requests.

As powerful as these AI capabilities are: AI requires equally powerful IT infrastructure, especially for the critical task of model training. Many businesses consume AI via cloud-based AI services such as ChatGPT, or adapt pre-trained AI models to their needs using RAG (retrieval-augmented generation). But large enterprises, service providers, and institutions often strive for a distinct business advantage from an individually designed and trained LLM (large language model). To achieve this, they prefer to perform their own in-house model training. So they need high-end servers designed with complex, demanding AI tasks in mind.

Comprehensive AI Portfolio

People familiar with AI know that HPE offers a comprehensive portfolio of AI infrastructure solutions and services. This portfolio ranges from starter kits for initial AI use case idea generation to enterprise-grade AI servers, high-performance computing (HPC) infrastructure, and even the most powerful supercomputers of the world.

For example, the Frontier supercomputer, deployed in 2021/22 at the Oak Ridge National Laboratory of the US Department of Energy, was designed and built by HPE. It is the world’s first so-called exascale computer, i.e. the first computer system capable of performing a billion times a billion floating-point operations per second.

New HPE ProLiant Compute XD Servers

Most enterprises have no need for world-class supercomputers. Instead, they need enterprise-grade equipment purpose-built for their AI needs. This is why HPE now broadens this AI portfolio with the new HPE ProLiant Compute XD server family, designed for large enterprises and service providers training their own AI models. To support these use cases, the servers are equipped with the latest AI chips from Intel, AMD, and NVIDIA. These are the two new family members:

* The HPE ProLiant Compute XD680 server is an air-cooled server specifically designed to support AI model training, tuning, and inference workloads with a focus on a good price-performance ratio. The compact chassis, only five rack unit high, houses eight Intel Gaudi 3 AI accelerators in a single compact node.

* The HPE ProLiant Compute XD685 server is built to accelerate the training of large, complex AI models, natural language processing, and multi-modal AI systems. One version, announced in October, features eight AMD Instinct MI325X accelerators and two AMD EPYC CPUs. Recently, HPE introduced a second version of the XD685 with NVIDIA GPUs. The server – which, like its XD680 sibling, comes in a compact five rack unit enclosure – is equipped with eight NVIDIA H200 SXM Tensor Core or NVIDIA Blackwell GPUs. In this case, GPUs, CPUs, and switches are fully liquid cooled for maximum heat dissipation and eco-friendliness.

The HPE ProLiant Compute XD680 server with Intel Gaudi 3 will be available in December 2024. The XD685 with NVIDIA HGX H200 will be available in January 2025, while the Blackwell version will be released at the same time as the NVIDIA Blackwell GPUs.

ProLiant Siblings with Cray DNA

These new ProLiant family members are "in-laws", so to speak: the XD servers have joined the ProLiant family coming from the Cray side of the HPE portfolio. With this, HPE acts on suggestions from its customer base who wished to see a clear distinction between HPE's enterprise AI portfolio and the supercomputing/HPC systems. Going forward, all the enterprise AI servers will bear the "ProLiant" family name, with the "XD" moniker indicating the Cray roots of a server. At the same time, supercomputing and HPC will continue to be the dedicated domain of Cray systems. This makes for a very neat division of labor.

However, this introduction of XD systems is not just a simple relabeling exercise. The new HPE ProLiant Compute XD680 and XD685 servers have their Cray DNA, but also some pure ProLiant heritage. They leverage a single, HPE-designed modular chassis, incorporate HPE ProLiant systems boards, and – most importantly – feature HPE’s Integrated Lights-Out (iLO) management solution.

HPE's iLO management allows authorized staff to remotely access the servers out-of-band. This means that administrators utilize a dedicated remote management connection, independently from the corporate LAN used for regular (in-band) networking connectivity. This considerably improves security and emergency accessibility compared to standard in-band remote access.

Getting up to Speed with HPE Services

To complement the new server offerings, HPE also offers a range of enterprise AI services. This allows customers to draw on HPE's decades of expertise in speedy deployment of the most complex AI infrastructures, providing for quick AI rollouts all over the world.

To summarize: The HPE ProLiant server family powers enterprise and service provider AI workloads from the smallest installation to the most complex environments. HPE provides customers with an exact fit for their AI infrastructure needs – along the whole AI lifecycle, from the initial demo use case to model training and finally inference of LLMs and multi-modal AI systems.

High-Performance Servers for AI Projects

Early adopters of the printing press or of the steam engine had to find out by trial and error which machinery and which sizing was the best fit for their individual requirements. The new HPE ProLiant Compute XD680 and XD 685 servers allow customers to skip this trial and error phase: the new servers combine Cray DNA with HPE ProLiant expertise to bring enterprises and service providers cost-effective and high-performance solutions for their AI model training and inference.

For more information on the new HPE ProLiant Compute XD680 and XD685 servers, check HPE webpage and watch these two explainer videos. If you want to discuss this topic in more detail, simply contact me via email.