Everpure chief technology officer Rob Lee put Nvidia at the core of the vendor’s AI strategy this week, saying it was seeing a number of traditional HPC use cases switching to accelerated computing. Lee was speaking at the vendor’s Accelerate Conference in Las Vegas this week, where chief executive officer Charlie Giancarlo outlined the storage vendor’s newly minted “data primacy” strategy.
This includes Everpure Data Intelligence, built on its recent 1touch.io acquisition. The firm pitched this as discovering, classifying and contextualizing enterprise information at source. That includes data on Everpure’s own flash-based kit, as well as third party storage and public clouds and SaaS. It claims it will deliver universal discovery, automated governance, and AI-ready context.
Giancarlo said this would enable companies to circumvent the problem of data being defined by applications.
“AI is going to these systems to try to come up with an answer to something,” he said. “But the data inside is different for each one of them, how does AI make sense of that?”
It also rolled out updates to the Enterprise Data Cloud platform it debuted last year. These include Evergreen//One Overdrive, which provides a cloud-like performance boost to handle spikes in on-prem storage up to 25 percent, without permanent subscription updates.
New features for the Everpure Control Plane include workload rebalance and mobility, due by the end of this year. Also slated for later this year is Fusion compliance and agentic triage to tackle configuration drift with agentic AI suggestion root causes. Copilot workflow execution will give storage admins the ability to use natural language to manage their global estates. Meanwhile its Everpure Data Stream will, the company claimed, reduce data preparation for AI from months to minutes, and enforcing stream-level access controls within the corporate network. It is underpinned by the Nvidia AI Data Platform reference design.
Given that Everpure has announced price rises due to component shortages, and that Nvidia’s parts are both expensive and hard to procure, we wondered whether the company should be working more closely with other providers.
Lee said that the company was working with other GPU providers, including AMD. But, he continued: “The way I would put it is we're definitely leading our AI solutions, and I would say product orientation and product feature targeting based on our partnership with Nvidia, which is both deep and very broad across the portfolio.”
All the same time, he said, “Most of what we’re doing in that space can be adapted to work on other GPU platforms. I think that’s an area that will in many ways be dictated by customer and market demand.”
And, he added: “Nvidia is the dominant player in that space, not just because of market dynamics, but because they've been so effective at delivering solutions and getting customers to the results.”
Everyone wants choice, Lee added, but customers right now are more focused on “How do I get to ROI. How do I deploy.”
That was about more than just deploying infrastructure, he said but to “start making meaningful use of data, and then actually get to results. I think that's where the partnership with Nvidia has been so mutually beneficial.”
More broadly, he said, some traditional HPC use cases were making more use of accelerated computing. “I think we do see some inbound demand for solutions like Flashblade//Exa from that part of the market.”
He pointed to recent sales of the AI focused Flashblade//Exa into the financial service sector. He put this down to “quant traders, hedge funds that were using GPUs to accelerate model back testing. A lot of the types of use cases that historically might sit in a traditional HPC type solution, but now have been fully modernized to look much more like an AI environment.”
These customers were finding that “Recasting those jobs that work those algorithms into GPUs can now be so much more effective than on traditional HPC style compute, that it's worth doing. They're getting better and faster results, and then that has all of the associated infrastructure demands and impacts that come with it.”
One of the big demands that dogs AI installations is energy – and Nvidia’s parts use a lot of it. Everpure has long emphasized the power consumption advantage that comes with using flash compared to hard disk drives, while better data management – and less copies of the same data – can also trim power bills.
Lee accepted that power consumption was not always top of the list for companies racing to deploy AI, or the companies supplying them. But this would change once the current technology explosion settles, he said.
“Nobody is focused on optimization up front, they're just focused on being first to market, being ahead of the next guy.” When that rush plateaus out, he says, “That's when people start focusing on improvements, and it's almost always the case that you get significant, order of magnitude improvements that follow from that.”
And Lee said there was plenty of room for improvement in key areas, even as the industry starts to fret about reaching physical limits of existing semiconductor technology. Quantum computing advocates claim we have to adopt the technology because existing CMOS technologies are reaching their limits.
But, Lee said, there was plenty of runway to go. “If you look at Moore's Law, for example, we were saying that Moore's Law was dead for ten or fifteen years before we really started hitting a wall.”
“When the thing you were doing before stops producing incremental improvement results, the next step is you think about how do I do things differently? And history has shown us it usually takes a couple, four or five tries of that thing before you really kind of tap out.”
“Whether it's semiconductor, manufacturing tolerances, whether it's photonics, I think all of these spaces have a long way to go,” he argued. With the exception of one, “hard disk drives, like that thing has actually tapped out.”
As for quantum, Lee says “We’re thinking about it but more from the point of view of post quantum encryption. I think that's the most tangible, actionable thing you can do.”
The fact that most putative quantum architectures still require temperatures in the minus 270 degrees range suggests to Lee that “it's gonna be quite some time before quantum computing really hits any sort of production level.” And even then, “quantum is not something that's going to make back office IT systems any faster. Breaking encryption algorithms, yes. Now, how about the in between? I think that's something that will take a while to sort out.”