Discover 2023: HPE Boosts Its Native AI Architecture

During the European version of its Discover event, HPE focused its speech and announcements around AI. The vendor is creating a dedicated architecture for this technology, including developments in Greenlake’s offering in storage or capabilities to develop and test AI models. The partnership with Nvidia is also strengthened.

“Discover is AI,” said Antonio Neri, CEO of HPE, in the introduction to the supplier’s European event taking place in Barcelona. He notes that “the adoption of AI and generative AI is happening at a speed never seen before.” But he points out that “AI workloads need resources, intensive compute capabilities, optimized file storage and dedicated software, as well as integration support.” A layer that HPE covers under the term “native AI architecture”.

A multi-level architecture

Various services are offered within this strategy, particularly within Greenlake (the company’s IT-as-a-service offering). HPE has updated its Machine Learning Development Environment offering, a studio for training models. It integrates additional features such as fine-tuning (model optimization) and the possibility of creating prototypes and tests. Complete the Greenlake for LLM service presented last June during Discover in the United States. On the software side, the Ezmerald analytics environment also benefits from a functional upgrade, optimizing data lakes for use by GPU and CPU accelerators. Ezmeral now supports more third-party tools, such as Whylogs for model monitoring and Voltron Data for faster, GPU-accelerated data queries.

On the storage side, HPE extends the Greenlake for File Storage offering for AI processing. “Our customers have asked us to work in environments that accommodate more unstructured data, have greater density, and improve performance,” Patrick Osborne, manager of cloud data infrastructure in HPE’s Storage division, told us. For the record, HPE introduced Greenlake for File Storage last summer, which was based on Vast Data solution integration. Today, the effort is focused on Alletra MP full flash arrays. Density has been boosted by support for 30TB NVMe drives and a capacity that can range from 88 to 250 PB. On the performance side, HPE announces support for Nvidia’s Quantum-2 Infiniband switch optimized for GPU computations. Greenlake for File Storage for AI will be available in 2024.

An extension of the partnership with Nvidia

Just like the competition, HPE cannot claim to be a significant player in AI without partnering with Nvidia. But the company, now based in Houston, has one advantage over its competitors: HPC expertise. “AI needs intensive computing and HPE has an ideal position in this area”, recalls Antonio Neri. He takes the opportunity to specify that this activity grows annually by almost 40% compared to that of servers, which is in sharp decline in the fourth quarter of 2023.

Antonio Neri, CEO of HPE and Manuvir Das, vice president of enterprise business at Nvidia, presented the extension of the partnership between the two companies around AI. (Credit: JC)

Together with Manuvir Das, vice president of enterprise business at Nvidia, he presented different building blocks for enterprises for training, inferencing and deploying AI applications. Based on the ProLiant DL380a server, the platform comes pre-configured with L40S GPU accelerators, BlueField-3 DPUs and Spectrum-X (SmartNic Ethernet for the network part). Initially, it scales with 16 servers and 64 GPU accelerators to accommodate Meta’s LLM, Llama 2, equipped with 70 billion hyperparameters. On the software front, Nvidia offers its Nemo framework for training and deploying generative AI. However, HPE is looking at other framework work like that offered by Alep Alpha (where HPE participated in fundraising last November). One thing is certain, with this pre-packaged offering and the native AI architecture, Antonio Neri aims to offer companies “an AI factory to deploy AI applications on demand”.

The channel will gain AI maturity

This architecture and its different offers will be made available to the channel to answer its customers’ questions. “We will hold workshops for partners on AI use cases. We have to make them progress in maturity”, explains Paul d’Aleira, channel manager at HPE France. Work that is part of the Ready Vantage program revised and officially launched since 1er November and which aims to “highlight the expertise of our partners”.

Regarding forecasts for 2024, Paulo d’Aleira is convinced that “growth will be driven by AI” and that the market will be more selective “more specialized servers, need for optimized storage with Alletra MP and data services”. And maturity will reach all companies. “AI is not just reserved for large accounts, we have several partners who target micro-enterprises and they want to be able to offer accessible solutions to this audience, without worrying about the infrastructure layer”, says the manager.

Leave a Comment