HPE’s Big Nvidia Vera CPU Bet: What You Need To Know

HPE announced it has outfitted its HPE ProLiant Compute DL394 Gen12 server with the Vera CPU—which is designed for agentic computing—to power HPE Private Cloud AI.

HPE is making a big Nvidia Vera CPU bet, bringing the eagerly awaited processor to HPE Private Cloud AI.

HPE said it has outfitted its HPE ProLiant Compute DL394 Gen12 server with the Vera CPU—which is designed for agentic computing—to power HPE Private Cloud AI.

The new Nvidia processor, which has shaken up the processor landscape, will be available in the fall.

The new processor comes with customers facing rising memory prices and supply constraints in the AI market.

Nvidia is promising that supply will not be an issue as it brings Vera to market.

The ARM-based Vera CPU lowers the cost for inference computing by eliminating CPU bottlenecks to inference computing, said John Carter, vice president of mainstream compute at HPE.

“The analogy I have been using with a lot of people is a traditional x86 processor is like a two-lane highway—when you get into a traffic jam it really doesn’t matter if you increase the speed limit,” said Carter. “You can increase the speed limit from 40 to 80, but the cars still aren’t moving. What we have really done here is opened up the traffic by adding more lanes to the highway. So it’s less about speed improvements. It is all about memory coherency and the removal of complexities like NUMA [Non Uniform Memory Access]. So you’ve got massive performance gains in specific kinds of workloads.”

Carter said a big differentiator for Vera is Nvidia’s focus on the specific workloads that will benefit from the ARM architecture.

“I really, really, like that Nvidia stayed super specific about exactly what workloads they were going after and did not try to attack the whole market,” he said. “That is where we have been unsuccessful as an industry with ARM. We don’t have to solve all those other problems. This is really focused on agentic, memory bottlenecks and bandwidth bottleneck problems.”

Higher Performance Throughput Than X86

The new Vera CPU provides 80 percent more performance on a per-core basis for throughput than x86 CPUs, said Nvidia Vice President of Enterprise Platforms and Solutions Chris Marriott.

“We’re really specific in the workload that we are going after here,” said Marriott. “Just like we do with our GPUs, we do the same thing with our CPUs and apply that goodness to these specific markets.”

Key to the performance is pairing the Vera CPU with low-powered LPDDR5X memory, said Marriott.

Nvidia is building ARM-based CPUs from laptops to rack-scale systems, he said.

“Literally we have products for every segment of the market by the end of this year to move the entire ecosystem [to ARM],” said Marriott. “That’s what makes me excited.”

Marriott said having HPE adopt Vera is a feather in Nvidia’s cap. “HPE is kind of the cornerstone of the enterprise market,” he said. “HPE has been serving HPC [high-performance compute] for 30 years and also uses a dense Vera-based liquid-cooled solution. It is kind of a match made in perfection.”

A Price-Competitive Solution For Inference Computing

Carter promised that the Vera-based HPE ProLiant Compute DL394 Gen12 will be priced competitively with x86 systems.

“When you think of a standard x86 system, this is going to be very competitive,” he said. “This is not a massive eight-GPU cluster with a price tag that something only the largest of customers can afford. When you think about a standard enterprise server, it is going to fall right in those typical ranges. The vast majority of this is standard enterprise components. It is not magic. It is taking everything that we already do for the enterprise and adding in some of the Nvidia special sauce.”

Marriott said Vera provides accelerated CPU performance but is right in the “price band” for inference workloads. “It is not super exotic,” he said. “The supply chain is there. It is using standard products and sheet metal.”

Nvidia has been determined to keep its Vera march into the market “simple” with a single Vera 88-core CPU for the inference market, said Marriott.

“We don’t have a SKU range of 30 different SKUs to go down into the entry-[level] standard x86 plain CPU [segment],” he said. “The simplicity of this allows HPE to build a single product SKU that goes directly into the mainstream to go attack these workloads. … The CPU and memory paired with it in a standard platform is going to be super competitive.”

HPE Provides iLO Security Silicon Root Of Trust With Vera

The Vera-based HPE ProLiant Compute DL394 Gen12 will support HPE’s long-standing “silicon root of trust” iLO technology, said Carter.

“What we realized here is the experience with which we build this technology really matters,” he said. “That is why we made it a central point that iLo is a central value of a ProLiant server and that was going to stay our management control point within a Vera-based system as well.”

Marriott said the iLO capability brings a powerful “single-pane-of- glass management and monitoring” for the huge HPE installed base of servers. “That’s super exciting,” he said. “That’s exactly what we wanted.”

Nvidia Is In Full Production, Ready To Provide Supply To HPE

Vera is in full production with full confidence in the supply chain to provide HPE processors, said Marriott.

“Our confidence in the platform is super solid,” he said. “It is here. It is coming. There are no issues of timeline. At this moment at least there is nothing from a memory standpoint that will affect the ramp of this product. There is no uncertainty in the memory supply of this product. The biggest risk I’ll say in the short term is literally the demand of CPUs that us and HPE and the industry is seeing as this agentic rollout begins.”

Marriott said he sees no “supply issues” with Vera at this point in time. “Going into the future, the demand issue for CPUs is going to be the limiting factor,” he said. “Can you just go out and buy 1,000 of these any day that you want? We hope that to be the case.”

A Call To Action For Partners For Vera: Get On Board

Carter said his call to action for partners is to get on board with the HPE ProLiant Compute DL394 Gen12 for HPE Private Cloud AI.

“Don’t wait,” said Carter. “We all know AI is moving faster than any of us can keep up with, even Nvidia. The faster you reach out, the better. Have a conversation with your Nvidia rep or your HPE PBM [Partner Business Manager]. Get a time set up and let’s talk now because by the time this thing is live and shipping, it is too late. You want to be in front of it.”

Carter said Vera opens the door for enterprises to “truly” scale and adopt inference computing. “You are going to get a system that takes advantage of all the latest technology with performance optimized specifically for this workload, but you are consuming it in a package that you can afford and you know how to deal with,” he said. “The simplification there just makes this very different than anything enterprises have had to date.”

Ultimately, Marriott said the Vera-based ProLiant DL394 on Private Cloud AI is going to dramatically shorten the time of tasks with inference-based AI computing.

“What it really allows us to do is just accomplish more and have personal assistants and agents,” he said. “That is really the impact. This is the start of that era of, ‘Look at what agents can do automatically now.’”