Cloud Hosting

At its annual HPE Discover event in Barcelona Thursday, Hewlett Packard Enterprise (HPE) unveiled a range of new and enhanced hybrid cloud products -- including storage, development tools, and rack-scale hardware co-engineered with Nvidia -- aimed at easing the development and deployment of generative AI systems for enterprises.

"We think the next decade is going to require full stack thinking from hardware all the way up through the software layers, as organizations lean into the deployment of AI applications," Evan Sparks, HPE's vice president and general manager for AI Solutions and Supercomputing Cloud told a news conference. "These applications are extremely computationally intensive. They require data as a first-class input and lots of lots of data. And our view is that this new architecture will require more thinking in terms of scale-up processing; they need to do workloads that are hybrid by design."

The announcement came a day after HPE announced its fourth-quarter earnings in which revenue and profit were down on the whole and down for traditional categories of storage and computing but up for high-performance computing, AI, and intelligent edge. While overall revenue declined 7% year over year to $7.4 billion, and traditional compute sales for servers was down 31% to $2.6 billion, intelligent edge revenue jumped 41% to $1.4 billion and HPC and AI products were up 37% to $1.2 billion.

HPE GreenLake for File Storage, an all-flash data platform, is adapting to better support growing and evolving large-scale AI workloads, including those based on generative AI and large language models (LLMs). The platform has been upgraded for AI model training and tuning. The improvements include increased performance density, integration with Nvidias new Quantum-2 InfiniBand for better GPU-focused computing, and significantly boosted throughput and capacity.

HPE previously said it is launching a cloud-based service powered by its supercomputers to support advanced AI applications. Initially focusing on LLMs, HPE plans to extend this service to specialized AI applications in areas including climate modeling, healthcare, finance, manufacturing, and transportation. Part of the HPE GreenLake portfolio, these services will be distinct in being exclusively cloud-based, unlike other GreenLake offerings that are on-premises or hybrid.

HPE and Nvidia are collaborating to develop a computing hardware and software platform specifically for generative AI tuning and inferencing -- the process where an AI model makes predictions and decisions on news data -- in enterprise data centers. The idea is to make it easier for businesses, regardless of their size, to adapt and use AI. This new offering is designed to be user-friendly, allowing companies to customize AI models with private data and deploy them from local devices to cloud-based systems.

The hardware component of the platform is based on a rack-scale architecture, where computing components are managed at the level of an entire rack of servers as a single entity to optimize resources. It includes HPE ProLiant Compute DL380a servers incorporating Nvidia L40S GPUs and Nvidia BlueField-3 DPUs (data processing units).

HPE's Machine Learning Development Environment Software has been updated to a managed service as a cloud-based solution for AI model training. The service, also included in its new AI tuning platform, is intended to help businesses quickly and safely start or advance their generative AI projects. It simplifies the training process of AI models, supposedly making it faster to develop them, and its designed to adapt to future needs, reducing the strain on management and processing resources. Additionally, the software includes new features for generative AI to enable quick testing and prototyping of models.

The new AI platform also includes HPE Ezmeral Software, which has been updated to make it easier and quicker for businesses to handle data, analytics, and AI tasks. The software platform, which allows users to run cloud-native or non-cloud-native applications in containers, is designed to work smoothly across various cloud environments. Key updates include a more efficient hybrid data lakehouse, which is now better optimized for GPU and CPU usage, facilitating easier data management and analysis. Also, integration with HPE Machine Learning Development Environment Software is meant to improve model training and tuning. The software also features better GPU allocation management, enhancing performance across different workloads. Moreover, Ezmeral now supports more third-party tools, such as Whylogs for monitoring models and Voltron Data for faster, GPU-accelerated data queries.

Customers will be able to order the new AI platform and the various updated components in the first quarter of 2024.

HPE also announced a GreenLake Flex Solution for Digital Twin built on Nvidia OVX-certified HPE ProLiant Gen11 servers with Nvidia L40S GPUs. The solution is meant to allow users to create, simulate, and optimize virtual assets and processes. Its a scalable, multi-GPU framework that provides the necessary infrastructure, software, and services to leverage the benefits of industrial digitalization. The digital twin features AI-accelerated infrastructure and Nvidia Omniverse Enterprise, designed to enable rapid insights from all available data. The company says the solution combines a public clouds flexibility with a private clouds security.

A digital twin is like a virtual model of a real-life object or system. This technology, which started with individual items, now covers bigger things like buildings, factories, and whole cities. Some people even think it can be used for people and the way things are done, making the idea of digital twins much broader.

View post:
HPE goes all-in for AI with new hybrid cloud compute, storage products - Network World