HPE, Nvidia team up to offer 'turnkey' private cloud racks for keeping AI loads local

Everyone wants to be best buddies with Jensen Huang's GPU juggernaut

HPE Discover This year HPE Discover 2024 is all about AI hardware developed through an expanded partnership with chip heavyweight Nvidia.

"Generative AI holds immense potential for enterprise transformation, but the complexities of fragmented AI technology contain too many risks and barriers that hamper large-scale enterprise adoption," said HPE chief Antonio Neri. "HPE and Nvidia co-developed a turnkey private cloud for AI that will enable enterprises to focus their resources."

At its most basic, Nvidia AI Compute by HPE consists of a trio of new servers designed for a range of needs, from LLM consumers up through full service providers. The most basic model, the HPE ProLiant Compute DL384 Gen12, ships with an Nvidia GH200 NVL2, while the top-end HPE Cray XD670 ships with eight Nvidia H200 Tensor Core GPUs. The mid-line model can hold up to eight Nvidia H200 NVL GPUs, and select versions are available with direct liquid cooling as well.

"Never before have Nvidia and HPE integrated our technologies so deeply," Nvidia boss Jensen Huang gushed.

HPE also announced that its GreenLake cloud platform had achieved Nvidia DGX BasePOD certification and OVX storage validation, and the company also shared news of new GreenLake OpsRamp capabilities for AI infrastructure observability.

The big news: A private Nvidia AI cloud from HPE

Cutting-edge AI hardware and some new GreenLake validations are all well and good, but the big announcement Neri shared at the keynote was undoubtedly the new HPE Private Cloud AI offerings, which the company is marketing to customers as being able to get up and running "in three clicks."

Hardware offerings are included here too. An HPE Private Cloud AI setup includes one or more hardware racks in four tiers: A small one for doing AI inferencing, a medium tier that adds retrieval-augmented generation (RAG) capabilities, and a large and XL setup that adds various degrees of machine learning fine-tuning options.

Regardless of the tiers, which vary mostly in relation to the type and number of Nvidia chips packed into them, HPE Private Cloud AI hardware is based entirely on GreenLake Cloud with a private control plan layered on top, HPE said. Nvidia's entire AI enterprise suite is also included, as is its NIM inference microservices. 

Depending on how customers want to set them up, the racks can be entirely self-managed or handled by HPE. Each tier is able to operate as a standalone on-prem AI solution or as a hybrid model.

For the curious, HPE Private Cloud AI racks come with a minimum of four Nvidia L405 GPUs at the small end, and as many as 24 Nvidia GH200 NVL2 GPUs in fully specced-out, extra-large configuration.

As for what HPE Private Cloud offers over competitors, the enterprise biz said it doesn't think it has any competition when it comes to winning customers looking to invest in AI who want to avoid spending their entire IT budget on streaming data to a cloud provider for analysis.

"We have designed this to be a turnkey single click," HPE chief technology officer Fidelma Russo said during a press briefing last week, "so we do not believe there are similar offerings in the marketplace." 

In other words, HPE seems convinced it's come up with a product that's going to be perfect for corporates with a lot of data that are looking to invest in AI without having to move said data out of their own systems.

"Enterprises around the world are moving from a public-cloud first approach to a hybrid one," Russo said at last year's HPE security analyst meeting, suggesting HPE's strategy has been in the works for a while.

"They've come to the realization that as their data grows and becomes more distributed in nature, it is neither scalable nor economical to rely solely on the public cloud."

But industry experts aren't entirely sure some businesses will want to take on the other expense of managing their own AI workloads: The risk that comes with moving all that infrastructure on site. 

"HPE must use Discover to build a convincing story that enterprises should minimize risk by exerting full control over their AI initiatives," Informa Fellow Steve Brazier told The Register.

"[HPE] also needs to convince Nvidia to provide it with a sufficient supply of Blackwell chips," Brazier added. In other words, as optimistic as HPE makes its future look, there's still a lot of work for the company to do to prove it's a worthwhile alternative to all the cloud-based AI alternatives.

HPE plans to release Private Cloud AI this fall, with the same timing for all the hardware announced today as well, with the exception of the new Nvidia H200 NVL-powered Cray XD670, which is expected to be available this summer. ®

More about


Send us news

Other stories you might like