In The Arena by TechArena - How AI-Optimized Storage is Eliminating Data Bottlenecks

Episode Date: February 12, 2025

Join us on Data Insights as Mark Klarzynski from PEAK:AIO explores how high-performance AI storage is driving innovation in conservation, healthcare, and edge computing for a sustainable future.

Welcome to the Tech Arena, featuring authentic discussions between tech's leading innovators and our host, Allison Klein. Now let's step into the arena. Welcome to the Tech Arena Data Insights Podcast. I'm Alison Klein. And because it's a Data Insights Podcast, it means I'm joined by my co-host, Janice Narowski with SolidIME. Janice, welcome to the program.

Oh, thank you, Alison. It's great to be back. Janice, tell me who we're talking to today and what is our topic? We have another great topic. Today we have with us Mark Klosinski from Peak AIO. Peak AIO is an up and coming organization who I'm very passionate about. They do some really cool things. And so today we're going to talk with Mark who is their chief strategy officer and co-founder.

Welcome to the podcast, Mark. Hey, thank you, Janiece. Hey, Mark, Peak has not been on the show before. Do you want to just start with an introduction of Peak, AIO, and what you're delivering to the industry? I'm Mark Lejinsky, founder of Peak AIO. We specialize in high-performance storage built specifically for AI and GPU compute. Our focus is to ensure that AI workloads run at full performance, remove bottlenecks, reduce space, power and complexity, and allow customers to get the most out of their GPUs.

So Mark, Peak, as I mentioned a moment ago, does deliver really innovative solutions. And it couldn't be more important of a time than now because data is paramount and organizations are trying to figure how to maximize their data. Can you tell us a little bit about some of the customers you're working with? I mean, London Zoo is one of them and they are dealing with some very unique challenges, but can you tell us a little bit about your work with the zoo? Yeah, the zoo specifically, the Zoological Society of London, ZSL as we call them, is using AI for some truly amazing conservation work, which is actually making a real impact today and it's going to be exciting what they do tomorrow and going forward.

They do so many projects, maybe we just look at one example of their projects, tracking endangered species across remote and challenging environments, managing many thousands of camera tracks, collecting vast amounts of data and imagery from the field. Now, you have many different projects, whether that's hedgehogs in urban London as we've seen, or whether it's tigers in India, each with millions and millions of images and other data. So firstly, these images need cleaned up by a model that detects and filters out non-reliant items like humans or cars and park benches, so hence isolating the animals. Now, when you have data sets, you'd analyzing learning and training,

then you get to get insights. And in this case, we can look at, and we are simplifying this down to a couple of projects, but look at how does urbanization encroach on natural habitats and how a habitat fragmentates the animals and their natural movements. So if we look at hedgehogs, for example, they have over 50 million images of hedgehogs and through training that they've managed to track their movements, develop as the urbanization of London grows, develop green corridors that allows them to mix with other hedgehogs and continue evolution. If we talk about Indians tigers, I believe they have over 10,000 camera traps that are

taking multi-million images. And those images, once we find again, they track movements and allow them to make informed and predictive decisions on how to help protect the animals. Now here's only two of the projects that we're talking about, only two animals. But if you now consider even with those animals, the addition of DNA, in the past, as I was only able to look at small sections of DNA, now they can read the entire DNA structure. Again, these results in significant amounts of data usage. All of this data needs to be fed to GPUs to gain insights.

When I heard that we were going to be talking about hedgehogs and tigers on the show, I was so excited because these are fantastic examples of AI at the edge. And what is opening up for scientists in terms of technology adoption. One of the things that I think about in terms of the zoo is there must be some unique challenges in these environments to navigate, to deploy technology. What are those and how have you overcome them? Actually, that's a really important question and great observation because it's very true. Although some would say this is, but it is really, this is the London

Sioux and in their case, they've converted what was an office into a mini data center. And so if you think about that, it presents quite a lot of unique challenges. So environmental constraints, many edge deployments face power, cooling, space limitations. Again, this was an office. It was not designed for racks of supercomputers.

And GPUs demand the majority of this power, space, and of course, budget. So storage has to be compact. It has to be efficient, and it has to be durable. Now, when you consider the scale of the images we just mentioned, adding DNA sequencing, both from animal, e.g. hair and environmental, water, soil, plants, then all the elements such as latency and bandwidth, edge environments lack the luxury of the hyperscaler data centers. However, they still need AI models to process the data locally at maximum

performance within those power and space constraints environments. That's where PKI-O really comes in. Our high-performance AI-optimized storage allows SEDISL to handle AI workloads directly at the edge, ensuring that conservationists get immediate insights without waiting on slow data transfers. An example of the filtering that I mentioned before about removing human objects. Prior to this installation, I recall they were able to make about three images per minute.

Now it's in the thousands. So the ability to deliver strong AI insights cannot be constrained by the reality that those insights are having to be delivered in small constraint ports. That's fascinating, Mark. It makes me want to just jump in and ask, how is it what you're doing with your solution different from what your competitors are doing in the space? Is there any sort of secret sauce to your solution different from what your competitors are doing in the space? Is there any sort of secret sauce to your solution? And then my second part of the question is,

what is the value of storage to help with the solution? So the traditional view of the storage has been you need more performance or you need more scale, add more nodes. And that was fine until GPUs came along and demanded all the space and all the power and most of the budget. And so just running and scaling at the edge isn't about adding more computers to each node. It means that now we have to deliver all that power that you would traditionally

see in what we would consider 12U, six nodes, and deliver that into you in one node. Our original primary focus and development was to deliver six times the performance in a sixth of the space with a sixth of the power. Now that sounds witchcraft almost, but we achieved that. By achieving that, that in turn enables people like SetASL and thousands aside to be able to achieve realistic AI and GPU modeling and other, and managing consume within real life environments.

It also enables not just educators, but enables the new star to not compromise on storage, they don't need to spend a multimillion dollar budget just for storage now. They can start their project. And in answer to that second question, really without the evolution of angry and me, particularly solid dying. And I'm not saying that just because you're here, as you know, we've been working with you for a few years now. Without that evolution, that was not possible. And so if we had a secret source, I would say it was the reality that we took a step

back and said, we need, just like NBME stopped and said, hey, the traditional protocols that we use for storage are no longer good enough. We need to create a new one and let's call it NBME. We did the same for storage and said, Hey, the traditional storage stack is not good enough. This is really good NVMe technology out there. Why don't we start afresh using this technology and not just try and incorporate into an existing stack.

So it wasn't an incremental development, it was a new development. And although everybody says that and everybody wants to believe that, if we think about it as a market, AI and GPU, it's a completely different market. Prior to that, we had HPC and we had enterprise. It's neither of those. It's a completely different market. And really after spending some time investigating this and being fortunate enough to work with many of the pioneers in the industry to look at what they needed, and more importantly, what they didn't need,

we realized that this new market, why would we think an old storage technology would ever work in them? How did we think we could rebrand it, call it AI, and it would magically work? It was a new world. we could rebrand it, call it AI, and it would magically work? It was a new world. It is a new world, and it's emerging, and it's growing every day and going in different directions. And so, Solidine and NVMe dramatically changed our ability to deliver that performance, to drive the density down, and to be able to deliver that to the user in a cost-effective

space saving manner. Now take that one step further and look at the incredible work that Solid9 have done over the last two years. We've gone from two years ago that within one node you would have had a few hundred terabytes to now petabytes and your 122 that you announced late last year. The market's absolutely really looking forward to the full supply of this. The impact of it will be gigantic because it now means that those that were struggling to house this level of data to develop more camera traps, to make those no longer just tigers, lions, hedgehogs, but now let's

talk about rivers, streams, natural landscapes. To be able to develop that workload and develop those sequences, they need to be stored and they need to be stored on something that can give them the performance that AI needs. And I'd say SolidEye stand alone on that. Now when you think about the amount of imagery that you're capturing and you think about the size of data, you know, one of the things that I just went to is that SolidEye just introduced the new 122 terabyte drives, delivering a different level of storage density than the world has ever seen. How has Pica.io responded to that? And what do you think the opportunity is in terms of the types of practical

applications, like what we're talking about today, in terms of the ability to deal with more data? The industry and ours response is really strong. AI edge computing or AI computing and large scale analytics are driving demand for density, high performance flash storage. The 122 terabyte solid-I drive represents us hugely in this possibility and it brings unprecedented storage density to the AI driven environment in such a dense form. Now, it's bad that also in terms of power, we're not increasing power, but we're doubling the density.

So we really hooked onto this and as Denise knows, we've worked for SolidI now for a couple of years. We're actually developing additional technology to go alongside these drives that will enable us to actually decrease the power by another 50% because they are such a game changer, not just in terms of what they enable the customer to achieve, how much data that they can store, which is obviously important, but this is a world of data. And although maybe the days of big data are over, the days of good data are growing and good data is growing and now accessible.

And so being able to store, house, backup that within a off the shelf product in an off the shelf server with some software defined storage is a real game changer to the industry and the users. There's no need for them to learn anything significant, train, implement staff, administrate, it's there. It works and it's an amazing game changer for the industry. Yeah. Well said, Mark.

I couldn't agree more. I think what you guys are doing is a game changer for the industry and you are making it seamless, easy to deploy. And I love that you said that this is, I don't hear this often, right, that HPC is not necessarily AI. It's not one and the same, it's different. And we don't always hear that transparency.

So thank you for that. One of the other things I wanted to ask you is you gave a lot of examples of the stories you've been working on at the zoo, but can you tell us a little bit about the work you guys are doing in the healthcare field? Because that's where I see a lot of transformation possibilities with your solution. Yeah.

And if you think about it, it's a logical place for AI. Because one thing AI is good at is imagery. And so we have an MRI scan. It's very good at looking at the MRI scan. Unfortunately we don't have an abundance of radiographers. Certainly in the UK it's not uncommon to go for an MRI scan, a PET scan actually, and wait several weeks for that to be reviewed. And in the UK that's often needs to be reviewed by two radiographers.

So there's obvious advantages at every level from the patient, the outcome, the expenditure, the government costs, and the personnel costs for AI to take a stronger role in that. And we were fortunate enough when we started PKI-O to start with and alongside some of the medical pioneers because that's an obvious starting place for AI. It has all the benefits I mentioned at every level. And so we learned so much from these guys and being able to see what they are doing is tremendous.

Our systems are already with Solidine inside the National Health Service, the NHS in the UK, that are evaluating and continuing learning, but evaluating MRI scans, for instance, they can probably detect whether or not the brain scan that you've just had is problematic or not by the time you put your shoelaces on. Now the first thing is to say that's a bit scary, but look at it the other way. It doesn't actually stop a diagnosis by a human. All it does is it comes up with three outcomes for one scenario. One, this is bad and looks like not a good job.

I don't really know. Or this is good. Now, if it's bad, why bother waiting for a human six weeks to look at it? Why not send them straight to the consulting? And so you've saved six weeks of time for the individual, the patient and the healthcare system and likely a significant cost in terms of treatment and a better outcome. If it's good or maybe, then send it onto the human for a second view. At least you've reduced that burden. And as AI is becoming better and better and that feedback loop continues, then

that increases. And so healthcare, the PKIO is still quite a passion. We work quite heavily with an organization called LEHI in the UK, which is the London Institute of Healthcare Engineering and one of the pioneers of healthcare in the world. It wrote something called MONAI, which is considered by own videos like the operating system for healthcare. So globally, most healthcare professionals now are writing their AI models based upon that MonAI, which we unsolidized, we're actually part of the pioneering work they did.

We're in the original one AI development. Mark, this has been an incredible story and the use cases that you're unpacking have just really ignited my brain in terms of what the industry is doing with this powerful technology. One final question for you, where can folks go to find out more about

the PKAO solutions we talked about today and engage with your team? Clearly, there's pkio.com is a good start. We have people around the world. The contact form will work its way to someone local. One of the advantages of a name like Klozinski, you can find me on LinkedIn easily enough. So I'm reachable and I actually am fortunate that I maintain contact with users and customers because without elongating this much further,

the one thing that's different about PKI and AI is our passion to see it evolve and our acceptance that we are not driving it, that they are driving themselves and pioneering themselves. And the only way we can make good solutions is by working with them and understanding their needs, their different paths and their different challenges. And taking the amazing technology that SolidIde has created and making that usable with some of our secret sauce and abilities. So reach out to me, reach out to the webpage and ladies, it's been amazing to review it. This has been such a great episode of Data Insights.

Mark and Janice, thank you so much for being on the program today. It's a real pleasure. Thank you, Mark. And thank you, Allison. Thank you. Thanks for joining the Tech Arena. Subscribe and engage at our website, thetecharena.net. kid.