HPE partners with Nvidia to offer 'turnkey' GenAI development and deployment
Hewlett Packard Enterprise (HPE) has teamed up with Nvidia to offer what they are touting as an integrated "turnkey" solution for organizations looking to adopt generative artificial intelligence (GenAI), but are put off by the complexities of developing and managing such workloads.
Dubbed Nvidia AI Computing by HPE, the product and service portfolio encompasses co-developed AI applications and will see both companies jointly pitch and deliver solutions to customers. They will do so alongside channel partners that include Deloitte, Infosys, and Wipro.
Also: AI's employment impact: 86% of workers fear job losses, but here's some good news
The expansion of the HPE-Nvidia partnership, which has spanned decades, was announced during HPE president and CEO Antonio Neri's keynote at HPE Discover 2024, held at the Sphere in Las Vegas this week. He was joined on stage by Nvidia's founder and CEO Jensen Huang.
Neri noted that GenAI holds significant transformative power, but the complexities of fragmented AI technology come with too many risks that hinder large-scale business adoption. Rushing in to adopt can be costly, especially for a company's most priced asset -- its data, he said.
Huang added that there are three key components in AI, namely, large language models (LLMs), the computing resources to process these models and data. Therefore, companies will need a computing stack, a model stack, and a data stack. Each of these is complex to deploy and manage, he said.
The HPE-Nvidia partnership has worked to productize these models, tapping Nvidia's AI Enterprise software platform including Nvidia NIM inference microservices, and HPE AI Essentials software, which provides curated AI and data foundation tools alongside a centralized control pane.
The "turnkey" solution will allow organizations that do not have the time or expertise to bring together all the capabilities, including training models, to focus their resources instead on developing new AI use cases, Neri said.
Key to this is the HPE Private Cloud AI, he said, which offers an integrated AI stack that comprises Nvidia Spectrum-X Ethernet networking, HPE GreenLake for file storage, and HPE ProLiant servers optimized to support Nvidia's L40S, H100 NVL Tensor Core GPUs, and GH200 NVL2 platform.
Also: Latest AI training benchmarks show Nvidia has no competition
AI requires a hybrid cloud by design to deliver GenAI effectively and through the full AI lifecycle, Neri said, echoing what he said in March at Nvidia GTC. "From training and tuning models on-premises, in a colocation facility or the public cloud, to inferencing at the edge, AI is a hybrid cloud workload," he said.
With the integrated HPE-Nvidia offering, Neri is pitching that users can get set up on their AI deployment in just three clicks and 24 seconds.
Huang said: "GenAI and accelerated computing are fueling a fundamental transformation as every industry races to join the industrial revolution. Never before have Nvidia and HPE integrated our technologies so deeply -- combining the entire Nvidia AI computing stack along with HPE's private cloud technology."
Removing the complexities and disconnect
The joint solution brings together technologies and teams that are not necessarily integrated within organizations, said Joseph Yang, HPE's Asia-Pacific and India general manager of HPC and AI.
AI teams (in companies that have them) typically run independently from the IT teams and may not even report to IT, said Yang in an interview with ZDNET on the sidelines of HPE Discover. They know how to build and train AI models, while IT teams are familiar with cloud architectures that host general-purpose workloads and may not understand AI infrastructures.
Also: Generative AI's biggest challenge is showing the ROI - here's why
There is a disconnect between the two, he said, noting that AI and cloud infrastructures are distinctly different. Cloud workloads, for instance, tend to be small, with one server able to host several virtual machines. In comparison, AI inferencing workloads are large, and running AI models requires significantly larger infrastructures, making these architectures complicated to manage.
IT teams also face growing pressure from management to adopt AI, further adding to the pressure and complexity of deploying GenAI, Yang said.
He added that organizations must decide what architecture they need to move forward with their AI plans, as their existing hardware infrastructure is a hodgepodge of servers that may be obsolete. And because they may not have invested in a private cloud or server farm to run AI workloads, they face limitations on what they can do since their existing environment is not scalable.
"Enterprises will need the right computing infrastructure and capabilities that enable them to accelerate innovation while minimizing complexities and risks associated with GenAI," Yang said. "The Nvidia AI Computing by HPE portfolio will empower enterprises to accelerate time to value with GenAI to drive new opportunities and growth."
Also: AI skills or AI-enhanced skills? What employers need could depend on you
Neri further noted that the private cloud deployment also will address concerns organizations may have about data security and sovereignty.
He added that HPE observes all local regulations and compliance requirements, so AI principles and policies will be applied according to local market needs.
According to HPE, the private cloud AI offering provides support for inference, fine-tuning, and RAG (retrieval-augmented generation) AI workloads that tap proprietary data, as well as controls for data privacy, security, and compliance. It also offers cloud ITOps and AIOps capabilities.
Powered by HPE GreenLake cloud services, the private cloud AI offering will allow businesses to automate and orchestrate endpoints, workloads, and data across hybrid environments.
Also: How my 4 favorite AI tools help me get more done at work
HPE Private Cloud AI is slated for general availability in the fall, alongside HPE ProLiant DL380a Gen12 server with Nvidia H200 NVL Tensor Core GPUs and HPE ProLiant DL384 Gen12 server with dual Nvidia GH200 NVL2.
HPE Cray XD670 server with Nvidia H200 NVL is scheduled for general availability in the summer.
Eileen Yu reported for ZDNET from HPE Discover 2024 in Las Vegas, at the invitation of Hewlett Packard Enterprise.