UT08x07: Building Servers for the Edge with Antillion

No matter how we define the edge, the special requirements for use in harsh environments drive unique product decisions. This episode of Utilizing Tech brought to you by soddy features Alistair Brad book, founder of Anion, discussing Edge Servers with Janice Roski and myself. It pays to start with the outcome to think about the constraints and to build solutions around that.

Welcome to Utilizing Tech, the podcast about emerging technology from Tech Field Day part of the Futureum Group. This season is presented by solid and focuses on new technology like AI at the Edge. I'm your host, Steven fst, organizer of the Tech Field Day event series.

And joining me today as my co-host is my old friend, Janice Roski. Welcome to the show. Thank you, Steven.

It's a pleasure to be back. Always, Always. It's been a lot of fun.

You know, we've done a couple of seasons here. We've had you on this podcast and the others, and every time you bring in somebody interesting. Now the most interesting thing that we're doing this season is we're trying to bring in folks who are actually, um, out there doing the work, implementing, making this stuff happen, you know?

Exactly. And what I'm excited, I'm excited about this series and this episode in particular, because you're right, we do bring in some cool partners. However, it's not too often that you get someone that can talk about the edge in a unique and different way.

And, um, I'm super excited about Alistair. I'm gonna let, I'm gonna give it over to him in just a moment too, to explain exactly what he does and who he is for the company. But Alistair and his team with Anion are building a really unique edge use case that can be used across many different industries.

And their vision and their focus about how they get work done and how they work with their partners, I think is incredibly different. So we're gonna learn a lot about Anion today. We're gonna lot learn a lot about the edge use cases, and then we're gonna find out, you know, where does AI and storage kind of fit into all this?

So, Well, welcome to the show. Alistair, why don't you tell us a little bit about yourself? Hey, yeah, great.

Thanks Sam, Steven and Janice for Invite me to. Yeah, um, my name's Alistair Bradick. I'm, uh, the founder of Anion.

Uh, I've been going about 10 years now. Um, background of Edge, we can talk about what Edge means. It's an interesting concept of what we all mean by edge.

But, uh, doing this for 20, 30 years, actually exploring how we collect and manage data and deliver capability to different types of both in the healthcare industry and, and more recently in defense. So yeah, that's kind of me and intrigued to see where we end up today talking around those sort of topics. So yeah, let's dive into that, um, because I, I think Edge is a really interesting topic and everybody that we've spoken to, um, within this series and also outsider, different definitions of, of what Edge is.

So I don't know, Alistair, can you tell us a little bit about what your viewpoint of Edge is and how do you define it with your company? Yeah, that's really interesting and probably whatever I define will be different to everybody else. I been, I everybody has their own view of what Edge is and probably it probably doesn't really matter to some extent.

It probably matters to the company who are designing things and the consumers who are going to use it to some extent. You just need a natural language that works between you and those customers that you both at least understand or at least can communicate on to us that edge. Well, it could be somebody jumping out of a plane, it could be somebody bo leading a boat and, and walking through water.

So my background at the moment is all to do a defense work. So when we talk about Edge, we actually have called it deployed for many years. So the Edge is more of a commercial construct that's come across more recently.

But prior to that, we've been talking about deployed comms within the defense industry for a long time because they have the ability, they all need to take it with them. They have to do stuff in these remote locations and they often don't have any comms. They don't have any power.

Uh, they don't have any of the niceties that you get in the data center or sitting at home now where I'm sitting here and I have all of the benefit of a, of a great broadband connection and great it. They have none of that. So I think that's the challenge for us is that the edge can be right at that, that that tactical space that where effectively what you have with you is all you have.

And then you've got phases leading up to that. And, and then the defense world as well as I suppose in in telecoms and other industries, you've got scaling of people and there's people scale then the types of technologies they need naturally scale with them. And that eventually, I suppose we say we're kind of not at the edge now, but I don't know if there's a really clear definition of when am I not at the edge.

I mean, I'm, are we at the edge now? I mean all of us are connected to a, a data center somewhere, which I, you who knows. So I think it's an interesting concept of what defines the edge.

I'm not sure that or a does it need to be defined? I don't know. Maybe for me it does because we build hardware that we say it's edge hardware.

So I suppose I need to define it, but I suppose as a consumer, you just want it to do what it does for the outcome that you are trying to get to at that stage. So maybe it doesn't matter so much, but yeah, it is inter it's something I pose every day, but we, we use the word edge, but I, I can't define it as well as probably I should do. Hence I've just muddled my way through that answer to try and explain where I think this edge could be.

Um, I suppose, Well, that, that's what I was kind of trying to get at with the, um, in the introduction. I, I'll say this like edge is as edge does, and um, the, the reason that it's interesting to talk to you about this is because, because you're not trying to come at it and say like, this is the box, this is where we need to put things into. You're trying to come at it and say, what, what do you need?

What kind of servers do you need? What are they, you know, what, what kind of specifications? What kind of, uh, configurations and, and, and by kind of coming at it based on the customer demand, rather than coming at it based on preconceived notions of what what it is I think has led to the development of what you've, what you've got.

So maybe it would help, do you want to talk a little bit, um, to, to kind of kick things off about the, uh, uh, servers at the edge that you've got? Yeah, let, let, let's go back and explain where we sort of come from, where we are now. And the reason where we are now, I suppose, is so traditionally in the IT world and my journey with defense in it, in that mix.

And we can go back prior to that with, with, with my work in the pharmaceutical world where we would kind to collect data for phase one, phase two, phase three, phase four trials. Which actually is interesting because a phase one trial is, is essentially a bedside trial. You, you are in a very small group of patients.

First, first dosage of, of a drug to a human is at phase one. You hope at that stage you are hoping you are just going to get some effects, but you don't wanna hurt anybody at that stage. And then as you go forward to phase four, you are looking at mass trials on a global basis.

And you can see the challenges there when it comes to collecting that data. We were looking at this 30 years ago of how do we collect data at each of those stages, but provide them back to the pharmaceuticals or the CROs to allow them effectively to submit that data to the FDA. So that's where it came from.

And then when, when I moved into defense, my first experience of that was actually the first and second Iraq wars, actually more the second Iraq war where we provided the, the comms for the, uh, multinational force, uh, in the south of, of Iraq. And the challenge there was, again, we had this splintering of groups about what type of data and what type of comms they had, both from fixed headquarters to mobile, uh, areas, and obviously a coalition of 44 or so odd countries. It was, it was a, a very large coalition there.

And essentially what we saw at that stage, we were either carrying compact, I think it was compact service in those Dell weren't really even around when we were there. I think it was compact then we were either carrying those things on our back, tipping out the dust on the back of a Humvee to try and make them. And generally they did work.

And, and that was an interesting factor is that we realized that that actually commercial comms, even in a desert, in a sandstorm in high heat, you could actually coax its way through. It wasn't always a hundred percent and it wouldn't hit the, uh, it wouldn't hit the metrics that compact or Dell or HP would ask you to, but you could kind of get it to work through. So that was kind of an interesting thing that we learned.

We also learned that portability, and actually we'll probably come back to this topic. We talk a lot in our company about portability and survivability, which those two factors are actually quite interesting because you could just say something's light or heavy. You could say something's big or small, but actually can you, can you carry it?

Can you move it to the location you want with the means of the legis logistics you have at that stage that logistics may be you as a human. The logistics could be a, a C one 30 with a certain size of, of, um, of area you can put it onto. Because if you don't go in that area, they're not taking food or other things they need to take.

So the balance of portability is really important to us. 'cause we need to be able to get our equipment and our capability to those locations where it's, where it's important. So we came back and we looked and we assumed that other people would be doing this.

We weren't a hardware company at this stage. We were actually more into doing software. We were doing loads of stuff with, um, computer associates and BNC and that's where my background sort of was in that middle phase.

But, um, we came back and we realized that actually nobody else was doing this. We were either having very old comms that, that were very, very dated and very hard to use, which was the defense world, or we had the stuff that was designed for data centers and there wasn't really much in the middle. So working with DSGL, which is the, the UK research arm for the government and into the MOD, we said, well how about we take commercial kit stuff that say Facebook we're using or starting to use in those days.

How about we take some of this kit and see if we could make it work for defense? We, and it becomes a risk balance. The risk is, well, maybe it's not designed to work as long as some of the other kit, but actually the balance of that is it doesn't cost as much and it's easier and it's more powerful so we can do more with it.

So we explored these ideas assuming actually that we wouldn't carry on doing hardware for, for very long. We always assumed that, that we would, either somebody would get better at us or, or, or actually we would just go back and go back into our software services world. But a, we kind of enjoyed it.

I think we were okay at doing it. You know, we, we seem to have forged a a a a different type of approach within the UK and NATO and, and now into the DOD. And it started this vision for us to say, could we put data center technology in a ditch?

That kind of was our mantra for a long time. Could we take, and sometimes actually I have to say to even the people I talk to is that just 'cause we can, doesn't mean we should because yes, I could give you an H 100 in a ditch, but do you need an H 100 in a ditch? Because there are challenges that come with that.

I can give it to you, but you've gotta power it, you've got to call it. So actually it was interesting how we have to balance what potentially we can achieve versus actually what is logical for what they're trying to do. And that comes back to your point, Steven, about outcomes.

And we call, we talk about outcomes quite a lot because the outcome should really be the driving force in most IT decisions. You know, in the end you should be thinking about what you're trying to achieve before you've decided what you've engineered. However, the majority of people in technology are, it are either engineers or people who want to be engineers.

So they kind liked this idea of being involved in describing the physicality and the technical aspects of it. And they kind of forget about what they were trying to do. 'cause they designed this amazing when you're like, what were you doing?

Were you doing, oh yeah, I just wanted to send an email. Why did you design this server then that can send a thousand emails? You just wanted one email.

Why did we not design that? So it's, we we're constantly having to bridge that gap between outcomes and what we can do and make sure that people take the right, the right choices. And that hopefully means that they get the best value for money, the best use, the best portability, the best viable, all of those factors come into a decision.

And actually in the commercial world, you don't generally have to make too many of those decisions. Maybe cost comes into it and performance, but generally, you know, it's a 19 inch rack. It's gonna go in a, in a space that may be changing.

You know, I've listened to quite a lot of podcasts recently where we're talking about power consumption. There was a great one recently with A-A-P-K-I-O and soddy, um, talking about exactly that factor of, you know, we can't just ignore the amount of energy that is taking now to do things. I mean, there was a, a quote in that podcast I was listening to which talked about that the the amount of energy going into just doing some degree of l and s in the future could be more than India could just to do some of these, could be more than what a whole, the, the, that country would.

And that's kind of mind blowing, isn't it? That that that we're gonna use that much energy to achieve stuff. And I think in the end, hardware will be important.

So things like soy will absolutely be important to innovate. We will be important 'cause we need to be able to get the power to it correctly and call it correctly. But I also think the actual engineers who are writing the software are going to be critical.

'cause in the end, you know, we, they need to be more efficient and well, we all need to be more efficient, I suppose, in, in that, in that chain because not one person's gonna solve this. Um, the question is do we need another LLMI suppose it's a different question, but we are human and we will always want the next best greatest thing. So I think the, the chance of stifling innovation because of climate is unlikely.

So it will revolve around us having to innovate to ensure that as we go through that journey, we consume less power, I suppose, and generate less heat, um, going forward. Um, Amazing. I I really love how you connected what I was initially speaking about your, you, your, your company and your vision and how you look at technology differently and kind of put the human factor first, right?

Yeah. Because there's lots of innovation out in tech, right? I've been in it for a long time.

Everyone's looking to one up so and so and be the next best thing. And sometimes you lose sight and you over-engineer and you start selling things that folks don't even really need or you're, you're, you know, oversubscribing on performance or density or power, whatever, right? So I love that you're kind of keeping that people first, human first mentality and designing and optimizing your solution to be what is actually needed.

Um, and I think that's why you're great partners with us as well because that's really truly our vision and why we're just sticking to storage. Um, you mentioned a little bit about, you know, what the people need and what the people want, right? So we're talking a lot about edge, we're talking a lot about ai, that big buzzword and do people really know what they need, right?

Everyone's trying to figure out ai, they're trying to figure out the edge. So, uh, and you brought up power and cooling, you brought up a lot of good stuff, Alistair. So, um, we'd love to get your insight on what you're seeing, particularly with your solution, be it that you guys dabble software and hardware.

Where are you seeing AI and, and kind of the needs of, of your customer base and, and how are they integrating all this together? That's a, that's a great, let's unpick that question 'cause there's, there's a lot in that that, uh, which is really, really interesting and fascinating. So AI in the defense, I mean, we can see it playing out in Ukraine at the moment.

The Ukraine war is, is clearly the, the beginning of a new way of doing it. Drone warfare. But drones are driven a lot by IT systems behind the scenes and the way we consume that data.

So actually that is a game changer in terms of the way that, that we're gonna have to think about how that information is both collected. So sensor fusion is going to become a really big thing. We need to bring all of this information in to mobile, let's call 'em data centers.

Really big trucks consuming both video, audio, uh, electromagnetic, uh, human source, other types of, of information needs to all be fused in real time. And not only does it need to be then taking a model that maybe has been learned on A DGX sitting back somewhere or multiple dgx is no doubt, but it's gonna have to refine that model because actually warfare is never quite as logical as you would expect it to be. Yes, there is quite a lot that you would assume was gonna be part of the model, but you're gonna have to keep refining that model constantly.

And I think that, that, that's not just the, I mean every industry is gonna have the same challenge. I suppose the challenge we we have is that we've gotta condense that into, into an object that doesn't have a broadband connection. It's probably got satcom even though it's great now with, you know, starlink is a better than where we ever were before and it will continue to get better, but it's still not directly connected into these cloud ecosystems.

So a lot more of what we need has to come with us and we then need to disperse and share that in such a way that we don't have big bandwidth to share that within a battle space. So we're gonna have to work out ways of how do we share this evolving model from these different sensors in such a way that everybody can learn from that. An example of that would be that you have, maybe you have a, you have a tank, a tank's coming through a from a, from a, from an area of a forest and a drone picks it up and it will take a photo of that potentially it'll take a video of that and then another drone sees that tank.

Now if you haven't understood it's the same tank, 'cause the tank may look very similar frankly. So if you don't know it's the same tank, you now think you've got two tanks. So somehow the AI actually has got to be able to put very small markers and they're using technology around putting it down to the pixel level markers to understand the difference between these objects.

And ensure then as that information passes between the different sensor fusions, we know it's only one tank. Because if suddenly we've got two tanks, three tanks, four tanks, we think we've got a much bigger problem than one tank. So actually that's a really interesting example of where we've got to be able to get better.

Now that doesn't particularly need any additional learning. That type of technology is available, it's just, you know, using vision learning. But there will be additional pieces that need to be generated in real time during that exercise or that battle that you potentially haven't got the time to send back to your big DGX server back, you know, sitting somewhere.

'cause you haven't got, you can't move the data or you haven't got time to make those decisions. It's real time decisions we need to make. So I think that's interesting on, on that side.

I think what we're seeing though, like lots of people, I think most people think they need more GPU than they probably do, but I don't think that's unique to our, I think that's, that's, that's been great marketing from the likes of NVIDIA and people to tell us we all need these things. And actually I think we've all forgotten about maybe there are other ways the CPU is still in there and sometimes A CPU can be equally useful as a GPU for certain functions, not A GPU doesn't always make it better, you know. And I think actually over time some models are actually aligning themselves to say, well you know what, if you've got a really good CPU, we can take some of those calls and actually you don't now need a GPU to do certain functions.

So I think we'd need to educate ourselves as engineers, but also educate then our, our customers downstream to make it clear about when A GPU or A DPU or an MPU or any, any of these additional processing units are valid. And when actually keeping it simple is the better option. Because the simpler we keep it a it's easier to maintain, it should be cheaper to buy, it's easier to, uh, to, to buy upfront.

'cause you're not on a supply chain. It includes so many different components that, you know, if you're on A GPU, El Musk has probably bought it already and put it into one of his big data centers. So we're not gonna better get that for a long time.

And if your system depends on it, you can't then get that and deliver it out quick enough. So I think that education's gonna be important. And I think actually that the, the software engineers will start writing software that needs less GPU over time for certain functions.

Certain things absolutely need it. Those core big things that happen, you know, right at the beginning of a, of a, of a model learning. Absolutely, you know, you're gonna need your your your blackwells and all of that.

But I think some things will change. So we're definitely, education is key and I think that's, that's key for us as well actually. 'cause we can get sometimes a bit like, oh wow, this is for A GPU at this, it'd be great.

Woo, put it, let's just put AI in the title and, and you know, and you know, I think we need to also stop doing that a little bit collectively. We are seeing storage, storage is a big thing, hence we partner so well with solid land we storage is getting much more prevalent. You take that sense of fusion challenge earlier, the amount of information we need to store and, but not only store, we need to be able to write it and read it very quickly.

Now we've probably pushed our customer far quicker than they had expected. I mean, they've been on, you know, scuzzy Saturn, you know, the fact that we've moved into MVM E so all of our platforms are now only MVME, um, and generally QLC as well because we want high density, you know, so we want to get as much into our systems as we possibly can at such a high rate. So, you know, PCI four at the moment, five, you know, coming out, we'll, we'll keep moving forward the best we can to mean that when they do hit those challenges, then the stumbling block won't be the hardware that we, they've either got or they've bought into our ecosystem.

It will be the way the software's written to make sure they take the best they can of that. But we are seeing a demand for more, more storage in a smaller space. And actually that's always been our challenge, our challenge for many years leading up to the revolution with the E one s, the E one LEE three I suppose, but we're more of an E one, you know, we love the E one.

It fits our, our vision so much more than the a standard two and a half type concert. And we can talk about that, why that is the case. But, um, I mean we can now get with 120 terabyte in a half 19 inch system where we do a lot of half 19 inch, a half, half a rack size, we can get 10 of those E one Ls in there.

So we're over a petabyte of storage in, you know, in that it's, it's, it's, it's amazing what we can now do. And, and I imagine that we could probably sit here in 12 months time and we'll be even more. I mean, it doesn't seem that there's an end to the amount of, there obviously is an end like Moore's Law when it came to CPUs, but it, it seems that we're on this, we're on this journey that doesn't seem yet to be ending in terms of the amount of storage that the, the likes of solid lamp can now fit into the same physical object.

But actually what's more important to me, not only is it physically not changing, which is vital because we want everything to be as small and as portable as possible, but actually you don't increase the amount of power I need to drive it. In fact, sometimes you give me less power, which is something that you do as well as a MD. So if you look at the, the A MD processes, if you go through looking at the, the 7,000 and the, the second series, the third is the fourth and the fifth actually, when they increase my core count, I actually don't get a big penalty for for, for power, which is amazing to me because that means essentially I can give my customer downstream something more performant but not then say, just to let you know, now you need a power station will behind you to run it.

I can say actually you can do that with less power and I'm gonna give you more calls. So yeah, to us the a MD soine marriage where we use them is amazing because we tend to be able to go forward with capability but either stand still with power or even sometimes even go back. And power to us is heat and we need to get rid of that heat.

You know, we, we do air cooled systems like the traditional ones. We do them in a slightly different way, hence we do a lot of half 19 inch even in the air cold fabric. So you can build interesting stuff, but we also do a lot where we, we have no air cooling, so it's all conduction cord or liquid cord or motion cording technologies.

Um, and that's when we have a big challenge because we have gotta get rid of that heat. If, you know, the movement of heat through those mediums is, is is more challenging than it is if you just put a load of fans over it. You know, even though, even though fans are are not efficient, air is not an efficient medium for conducting heat.

You can still get away with just throwing a lot of fans at it and, and eventually the heat will go out. I mean, assuming the ambience alright, but when you literally condense it down. So we've been doing conduction calling with solid D drive ever since the E one first came out.

We, in the first week of us getting it, the first thing we did was rip the heat sink off it, put it inside one of our conduction core systems. Um, and in and in the defense world, there's a, there's a concept where you can pull the heat out through, uh, effectively called V-A-V-P-X system is where you, you effectively use, um, uh, uh, a way to clamp it in the copper pulls the heat out, and then that then is where the heat goes out. And, and we've used that method in our system, in our, uh, conduction systems for five years now, six years now.

Um, with amazing effect. I mean it, it'd be nice not to have to rip the whole thing to pieces, but it's the way it works for us. And, and means we can get in something about this, you know, about centimeter high.

We can get, you know, six disks in that, in that fabric. Um, and they can pull it out and they can do what they want with it then and, and it becomes effective. It's a bit like having an Nintendo cartridge.

They have a cartridge of their information, which then if they need to run very quickly, they can pull out the data and they can run, if they need to transport that then, and it's secret, they can transport in certain ways that is easier than if it's 50 disks that are just thrown in a, in a, on the floor. And you've then gotta work out, well, which disc went with which disc? You know, how am I gonna reboot my system with if I don't know.

We deal with a lot of that challenges. Uh, and the E one s has been, has been the enabler. We were using M twos before, we never really liked the m twos.

It, we liked the, we liked the, the, for the mechanical format, but hated the performance and the general, they were consumer really not really enterprise. E one obviously gave us the best of both ours fully enterprise and in a format that meant we could do the same job and E one L. Yeah, I mean that's just amazing, isn't it?

I mean the amount of storage on E one LI know the ruler for many years from Intel from previous years, you know, never quite made it. And I still don't know if the E one will become as prevalent as, um, as other 'cause it is a challenge because it's 300 and something deep, so it makes servers bigger and so forth. We do it differently though.

We don't see a, we don't see a server the inside of a server as a one dimensional space. Most, most companies steer it as, as a single, even in a two year system. It's just effectively the same plane but just higher objects.

We actually see it as a full three dimensional space. You know, we're manipulating objects at different planes and at different levels and at different locations to see how much can we cram into the smallest space. The density is another phrase we use a lot of.

We're we're trying to increase the density of everything in our systems. And that's not just storage with solid design, but it's, it's, it's, it's everything. It's the amount of power we could get to it.

It's the amount of cause we can get the amount of, uh, of memory. So density is, is critical. The other interesting thing, we we've, we've, we've been playing around with the concept of disaggregated systems for six, seven years plus.

Uh, it is become more of a big thing now, unfortunately, from our point of view, when we liked the idea, A, we weren't clever enough to do the inventions that needed to happen. We, we, you know, we're, we're good at our stuff, but we're nowhere near as clever as these amazing people who can do this, these inventions. So we had the ideas but couldn't deliver on it because the things didn't exist.

We worked with Fujitsu n EEC for a while on using light to move PCIE. So we did a really great piece of work where we were using, and SAM Tech had a, had a vision of using light to move PCIE so we could disaggregate the PCIE bus remotely. And what that meant is that we weren't limited then by the density of a single object.

We could increase our density across multiple objects and disaggregate the functions we needed. That's become a lot easier now because the likes of CXL has come along. I mean, I say it's easier, CXL is still quite complicated.

1. I don't think it's really become that proliferated as much as people wanted to. I think two is starting to probably deliver what most people have wanted it to.

But for us, the ability to disaggregate say memory, to increase the memory footprint, but without increasing the, the physicality of the single object and saying to a customer, if you do need more memory, yes there is a, there is an overhead of of a bigger subsystem, but you can make that choice when you need it. You don't have to compromise at the beginning. So the reason we did this was because we didn't want people to have to say, well my object needs to be this big because I might want a GPU and I may want more memory and I may want this.

We wanted to say, well if you don't know you need it, let's make it the small as possible. Unless they allow you to add those capabilities to only those systems at the moment you needed to deliver it. And that, that's really vital for our world because if they don't have to take it, why give it to them?

'cause if they take it, they've then gotta power it. They haven't gotta maintain it, they haven't even gotta look after it. They then also have to potentially buy the same systems with all of this expensive capability upfront when actually they may only need 10% of their fleet to be able to be upgraded to that level.

Now it probably sounds a bit stupid, we should just be saying, Hey, buy the most expensive system that kind of isn't us. We would much rather they bought the right system and had the ability to grow or even contract when they needed to. 'cause actually it's not like in a data center where you can just throw it at it.

They literally have to sometimes put it in their backpack and carry it. And if they don't have to carry it, why would I give it to them to force them to have to go down that route? So storage will be big for us though as well.

I mean, I know storage disaggregation has been possible for a long time MVME of fabrics and, but again, I think with CXL it becomes an interesting approach of a, of a, of a standardized approach for us. That disaggregation of everything on that sort of PCIE five and six bus we can start doing. For us it would be memory and storage be the, the two big things with maybe GP as well moving those.

So yeah, sorry, very long answer to a question you asked earlier, but anyway, Amazing. You've covered quite a lot of ground there actually, uh, um, I will, will point out that season four of utilizing tech focused on CXL and one of the things that we focused on in there is exactly what you're talking about. It wasn't about necessarily disaggregation as the goal is about right sizing as the goal and about flexibility and making systems that were not bound by the strict structures of the size of a dim or, you know, it's about making things right.

And to me, I think that's really kind of the summary of this whole conversation. It's, it's, it's interesting isn't it, that without constraints people tend to just expand like crazy, but with constraints. And that's what makes the edge interesting to me is that it's a constrained system.

You cannot go beyond this physical size or you can't go beyond this power footprint or you can't go beyond this cooling or, you know, we have to make sure that it's rugged or we have to make sure that it's able to, you know, to, to handle disrupted communications. Oh, and disrupted communications means we have to do processing locally and that means we have to and, and, and there's this whole chain of thought that happens when you are constrained. That doesn't happen when you're in sort of a data center or a cloud environment where you can have as much of anything at any time within reason.

I mean, but, but even now, I mean, look at ai, it's not even within reason. It's just, you know, you can have as much of whatever you want, you know? Yeah.

When it comes to the edge, and especially in the applications you're describing, these constraints make us more creative. They make us come up with clever, uh, novel uses for technology that we already have, like the CPUs that we already, that we already have and, and, and new technology as it comes out. It makes us see new possibilities.

And, and that's what's inspiring I think about this whole story is that, you know, we can do more with less if only we just tried. Would, would you agree with me about that? I would, yeah.

I think that the idea of constraint is amazing. I I love that actually that phraseology you've come up with actually. And I think that's what, what, what drives me actually.

I think that the idea of not having unlimited options and, and having some constraints actually makes it more, more interesting. I dunno why. Maybe that's a, that's a part of my psychology maybe, I don't know.

But that idea of having a, having restrictions and having to work around them is fascinating. And you think about where potentially we want to go as a, as a, as a world. I mean, not, not where in my lifetime obviously, or or any of us probably, but you know, when we look to go to to Mars or other places you think these challenges we just spoken about, those constraints are absolutely gonna be in there when we start going into planetary, then those challenges are gonna be there.

Even on this world, we are gonna hit these challenges. We know that. Not maybe within the lifetimes of any of us on this call, but there will always be.

I think there are gonna be more constraints. We may, I don't know where the peak of unconstrain ability is in this world of ai, but at some point we will, we will flip over it and, and we will be constrained by the amount of power and and so forth. And I think, um, yeah, I think edge, I think a lot of other industries could learn from edge.

Even those ones that at the moment are unconstrained, I think they could learn. Yeah, I couldn't agree more. Uh, Alistair, I feel like, um, and Steven, you said this too, like, is the industry really trying hard enough, right?

Have we really looked at, you know, cooling storage? Um, we're starting to, we've got some new designs that we just placed out onto the market and everyone's like, whoa, what is that? That's cold plate.

Wait, it's cooled on all four sides. It's not just with the liquid tube. Wait, it's not dunk dunked in anything.

It's not diabolic cooling. What is it? Right?

So it's like, I mean, to your point, Alistair, I think your company and your vision and what you're doing is so ahead of itself, right ahead of the time because you aren't just looking at cooling, the GPU, the CPU, you are looking at the overall architecture. You're looking, how do I make this more seamless, portable, simple. Um, and again, that human, that human first focus, which I personally, um, really, really appreciate.

So, uh, I can't wait to learn more. Um, I feel like we could, we could talk all day about this, um, and, and still be like really interesting. So I actually think we should, we should have you back on at a later date, Alistair, and, and talk more about, you know, the stuff you're doing with PKIO and some of the other organizations, right?

Because we talked a lot about military today, and we talked a little bit about, you know, the, the other use cases you, you guys work in. But, um, it would be interesting to give our audience an understanding of how you're doing this differently for others. So this has been phenomenal.

I really appreciate your insight. Good. It's been a great chat.

I've really enjoyed it. Really, really good. Thank you very much.

Yeah, thank, thanks. It's, and I have to say too, the, this whole story of constraints and, and, and, and incredible, uh, challenges and, and, and how people meet, meet challenges at the edge. I mean, this is what we've been talking about at Edge Field Day, um, you know, the whole time through.

Um, I, I just love this discussion. We've talked to a bunch of companies in this space that are doing things like this that are kind of rising to the challenges that are placed on them. Um, it's, it's, it's, it's incredible.

So, uh, thank you very much, Alistair, for your time. Thank you so much for joining us. Before we go, um, I, I'm sure that people are gonna wanna continue the conversation with you.

Uh, where can they connect with you? They can, they can find me on LinkedIn. I'm not, I'm, I'm not very good with the whole social thing.

Other people are, I'm not, I'm, I'm probably from that generation. I didn't quite embrace it enough as I should have. Um, but LinkedIn, you can find me there.

com and, and reach out to us. Uh, obviously I'm doing a few podcasts with, um, with Solid Lamb at the moment, which is Alien to me doing meetings, but I'm enjoying them. Actually, I, I'm not really into the self-promotion discussing thing, but I'm, I, I love these types of ideas of just talking, which is great.

So, yeah, you know, there are ways to get hold of me, but I'm not prevalent on the Twitters and the LinkedIn, you, the other players. Yeah, I'm not great on those areas, but you can get hold of me. Well, You, you wouldn't know it.

Um, uh, excellent, excellent conversation. Um, and, and Janice, uh, you know, thanks for joining us here again. Um, what's New with Soy.

Thank you for having me again, Steven, this is always fun. com/ai and you might see another sneak peek of what we're doing with Anion there as well. Excellent.

I would love to see that. Um, and as for me, uh, you know, we've been really enjoying this season of, uh, utilizing Tech. Uh, you will find us, uh, show notes, more episodes and so on.

com. Uh, you'll also find us in your favorite podcast applications. Uh, you'll find us on YouTube.

Um, we would love it if you would leave us a comment, leave us a rating, leave us a review. Uh, it really helps us to, uh, direct where we go with this in the future. Um, this PO podcast, as we noted, was brought to you by Soy, as well as Tech Field Day, which is part of the Futurum Group.

Uh, we can also be found on, uh, the socials. Uh, find us on X Twitter, blue sky mastodon at utilizing Tech. Thanks for joining.

Uh, we've got a new episode coming next week. Uh, listen to the whole season. Uh, we'll love to to hear from you.

Thanks for joining us, and we'll see you next week.