“AI is very complicated in enterprise environments,” stated Pure Storage CEO Charlie Giancarlo at a press event during Pure Accelerate this week. To address this from a storage perspective, the company is adding several new features, services and SLAs to its offerings today. A new version of Pure Fusion, Evergreen//One for AI and an upcoming certification for an Nvidia architecture stand out the most.
Pure Accelerate is a fundamentally different show this year than last year. Indeed, on the first day, we hardly heard anyone mention the words FlashArray or FlashBlade. The announcement of a new DirectFlash Module (DFM) of no less than 150TB, doesn’t even make the news. This announcement is more or less mentioned in passing by spokespeople, even though it is double the capacity of the DFM announced during last year’s show. It makes the gap with the highest capacity HDD a whole lot bigger than it already was.
Also read: Pure Storage wants to make hard drives permanently obsolete
It is clear that Pure is currently working on different things than last year. In itself, that is refreshing. Last year’s story that HDD had had its day and that flash will definitely win out is slowly but surely becoming reality. This is not only clear from the stories we hear from Pure. The growth in affordable QLC flash among Pure’s competitors is also a good indication that organizations replace HDDs by flash in more and more tiers. Mind you, the flash storage that other parties are putting on the market is based on SSDs not DFMs. That’s a substantial difference, but for the point we want to make here it doesn’t matter much. Flash is clearly growing fast and displacing HDDs.
So what is Pure Accelerate all about this year? AI of course, how could it not be? Pure has been working hard in many areas on innovations in this area. The main goal here is to make and keep the complexity that AI brings as invisible as possible.
Pure Fusion gets major update
One of this year’s biggest announcements is that Pure has developed a new version of Fusion. Pure Fusion is a “private cloud of storage,” in Giancarlo’s words. It sees all storage as a single storage pool, whether on-prem or in the cloud. We described it in 2021, when Pure Fusion first hit the market, as storage-as-code.
However, that first version from 2021 had its limitations. It only worked in greenfield environments. Pure hosted the intelligence behind it in the cloud. The new version is embedded in Pure’s arrays. That ensures that it is now also backwards compatible, something that really belongs in everything Pure does. After all, it wants to be and remain evergreen. That is now possible with Fusion. There were also customers who did not want to or were not allowed to use the cloud. They can now use it too.
It was certainly not an easy update that Pure gave to Fusion. It had to work hard to keep the Fusion control plane and the existing control plane on the arrays from getting in each other’s way. Pure did not want to introduce a new external control plane.
Pure Fusion itself is not necessarily an AI-related announcement. That is, even if organizations are not yet using AI workloads, Fusion is of interest. Of course, AI does bring storage with it, so indirectly, optimal provisioning and setting up storage does matter. Viewed this way, it does have to do with it.
Evergreen//One for AI
Whereas Pure Fusion may not be explicitly an AI innovation, we can clearly say that of Evergreen//One for AI. This is a new Storage-as-a-Service (StaaS) offering from Pure. It aims to set up the storage needs of AI workloads for organizations without them having to continuously manage the storage themselves. That is, Pure gives performance guarantees. It bases these guarantees on the maximum bandwidth requirements of the GPUs. That should ensure that there are never any bottlenecks.
Pure developed Evergreen//One for AI for environments with large numbers of GPUs. There you have broadly two kinds of data. The first kind is working, the other is resting(data at work and data at rest, as it is called). The idea is that the data doing nothing also uses less expensive storage. However, it is not easy to predict how much of each kind of data is needed in the kind of large environments we are talking about here. So that’s what Evergreen//One for AI solves, according to Pure. It tracks actual usage and only charges for what customers use.
Nvidia SuperPOD certification and AI Copilot
To market a StaaS offering such as Evergreen//One for AI, it is obviously also important that organizations can assume that Pure knows what it is talking about when it comes to AI. For that, certifications from Nvidia are important. For Nvidia OVX and DGX BasePOD, these are already available. Later this year, however, the company expects to obtain certification for Nvidia DGX SuperPOD as well. This means that Pure’s design has been validated and performance guarantees can be given. It also indicates that upgrades don’t have a major operational impact. As usual in 2024, which we called the year of Ethernet earlier this year, Pure’s design is Ethernet-based.
Finally, as a final AI development at Pure, we would like to touch on the AI Copilot that the company has developed. According to Shawn Hansen, the GM of Pure’s Core Platform Business Unit, this is a “radically new way to manage and protect data.” It gives admins more insights and enables them to solve issues much quicker. “It makes an average admin good and a good admin great,” he states. Thanks to an NLP interface, you can ask questions of it in natural language and, if all goes well, you’ll quickly get to the answer that helps you along.
Pure’s AI Copilot uses the metadata the company collects from customers every year. This quickly adds up to about 25PB a year. This data covers not only the storage itself, but also how organizations deploy and use it. In addition, there are links to Pure’s knowledge base as well. The AI Copilot uses GPT4 from OpenAI to extract the actual results from all this data. Also interesting is that the output of the AI Copilot will not only be text, but also things like pictures of cables, maps, GUI and the like.
All in all, Pure is taking several steps today to address the complexity posed by AI. It is addressing both sides of the challenge. There are improvements to the storage offerings on which AI must run, but it also does not forget to continue to leverage AI to make storage easier to deploy and manage. Organizations need both components to put AI and storage where it can make a difference. After all the hardware improvements over the past few years at Pure, it’s the software that plays a crucial role in storage now. The company is very well aware of this and is playing to it extremely well this year.
