HPE: AI stack with models, servers, network, data management and admin console
HPE is also increasingly relying on a cooperation with Nvidia for AI - like many of its competitors. How can the provider set itself apart from them?
(Image: HPE)
- Harald Weiss
After many large system manufacturers have already announced extensive collaborations with Nvidia, HPE has now also presented a far-reaching collaboration with the AI chip leader at its latest Discover customer event. Under the name Nvidia AI Computing by HPE, a whole bundle of coordinated hardware and software was presented that is designed to make the development and operation of AI infrastructures faster and easier - a kind of hyper-converged AI infrastructure. The joint offering is marketed as an AI stack.
Nvidia CEO Jensen Huang explained its approach in the keynote speech: "AI essentially consists of three parts: the model, the computer technology and the data. The latter is the most important, because it is the only way a company can differentiate itself from the competition." However, the data area in particular is a problem child of AI - and not just the amount of data required to fine-tune a model, but above all the quality and compliance with all data protection requirements.
Different components are used for each individual stack area. The model stack essentially consists of an API platform via which preselected public LLMs can be integrated and customized. The Nvidia inference microservice is available for operating the LLMs. Nvidia's AI Enterprise, which also includes Nvidia's NIM, is used for the data area. According to Huang, this is used for vectorization, semantic embedding and data retrieval. "AI Enterprise accelerates and optimizes the development and deployment of production-grade copilots and other GenAI applications; it provides user-friendly microservices for optimized AI model inference, enabling a smooth transition from prototype to safe production environment," Huang said of this component.
On the HPE side, their AI Essentials software is used. It includes control and management functions designed to ensure that all AI processes are data protection-compliant and transparent. The compute area relies on HPE Greenlake for file storage and ProLiant servers with support for Nvidia L40S, Nvidia H100 NVL Tensor Core GPUs and Nvidia's GH200-NVL2 platform. Nvidia's Spectrum-X Ethernet is added for networking. All of this can be set up in a private cloud on premises, but a connection to the Greenlake cloud is required for orchestration. Here, HPE provides a control layer from whose console administrators can manage and monitor all components and end devices of the AI stack. Automation tools are also provided to reduce their workload.
But how does HPE want to set itself apart from the many competitors? Gartner analyst Julia Palmer says: "Nvidia AI Computing by HPE is an integrated, coordinated technology stack that unites all components under a single management umbrella. That's not easy when you consider how demanding the AI workloads are and what needs to be integrated." In addition, HPE partners Deloitte, HCLTech, Infosys, TCS and WIPRO have already announced that they will use the new AI stack to develop special AI applications. More information on the cooperation and the products can be found in the announcement.
HPE and Aleph Alpha: first major order
There was also an update at Discover on HPE's cooperation with Aleph Alpha and its Luminous model , which was announced a year ago. This included a major contract from a US federal agency to implement generative AI to analyze, consolidate and generate documents that are critical to national security. "To support these requirements, we implemented a private local LLM environment based on HPE supercomputing and AI technology, which the agency uses for training, fine-tuning and inference based on its own documents and databases," says Jonas Andrulis, CEO and founder of Aleph Alpha about this project. "In this way, the authority benefits from the general functions of the Luminous model, while confidential data remains completely private," Andrulis continues. Analysts also praise the HPE Aleph Alpha offering, particularly for its energy efficiency. "The horrendous computing costs are a big problem for all companies that want to use GenAI. With Aleph Alpha, HPE has chosen an LLM that is inherently more efficient than many other available models," said Mark Beccue, of Futurum Group, praising the Heidelberg-based company's program.
(anw)