@HPC Podcast Archives - OrionX.net - @HPCpodcast-71: Alain Andreoli – Industry View

Episode Date: October 24, 2023

We are starting a new feature, looking at HPC, AI, and other advanced technologies through the lens of industry leaders. In this episode, we have the pleasure of a very lively conversation with Alain Andreoli, a longtime luminary of HPC and IT. Mr. Andreoli was with HPE for more than 7 year where he served as group president and EVP of the Hybrid IT Group, helping shape HPE's strategy for HPC including the acquisition of SGI in 2017. [audio mp3="https://orionx.net/wp-content/uploads/2023/10/071@HPCpodcast_Alain-Andreoli_20231024.mp3"][/audio] The post @HPCpodcast-71: Alain Andreoli – Industry View appeared first on OrionX.net.

HPC moved from a niche to becoming the core of the market. It's a fascinating business because beyond what an individual developer or customer would need, there is this geopolitical dimension and quantum will take it to the next level eventually. Customers need to have people they can talk to who know what they're talking about and can be anchored in their thinking and be a sounding ball in this first phase of what to build to make my company better in AI. Is it going to help us be better humans or be humans longer? Or is it going to make us completely irrelevant, completely obsolete, and we are like, that's it.

From OrionX in association with InsideHPC, this is the AtHPC podcast. Join Shaheen Khan and Doug Black as they discuss supercomputing technologies and the applications, markets, and policies that shape them. Thank you for being with us. Hi everyone, welcome to the AtHPC podcast. Great to be with you again, Shaheen. Great to be here and our very special guest today as well. Yeah, we have with us Alain Andreoli, longtime luminary in HPC and IT. Alain was with HPE for more than seven years and served as group president and EVP of the hybrid IT group. And during that time, we understand he was quite instrumental in HPE's acquisition of SGI

in 2017. Earlier, he was at Sun Microsystems. He was president of the European operations, and he was a senior executive at NTT, Oracle, and Texas Instruments, among other organizations. Alain, welcome. Thank you, Doug and Shaheen. It's a pleasure to be here. Okay, very good. So I guess we could start with a pretty broad question. You know, with the rapid advancements in supercomputing over the past 30 years, what are some of the key breakthroughs or milestones, in your view, that you've seen? I think the outcome of a lot of milestones is that HPC moved from a niche to becoming the core of the market. If you think about the reason why people still have IT on-prem and not on the cloud, it's

mostly with HPC. And tomorrow, it might be because of AI as well. So it may be the second wave of the rebirth of on-prem. Now, when I joined HP in 2013, you may remember Meg Whitman was at the helm and trying to reinvent the company. And Meg and Antonio Neri, who was leading the server business at the time, really had a dilemma in front of them,

looking at what's going to happen to the compute and store business 10 years down the road. How can we organize for the future? And I was hired to try to help resolving this dilemma and equation. And we started with two dimensions. And this was 10 years ago, right? 2013. We felt there were two segments of the market that would be pivotal to the future.

One was the cloud. The other one was HPC. So my first job was be the HPC and cloud guy and try to help us understand where to go. I will pass on the cloud because that's not the meeting we're having now. But for HPC, it was basically, we concluded very quickly that it would be a different way of doing business. There was a lot of government business initially. Most of the business was on a few segments and customers needed a touch that went far beyond the catalog product that we had at the time, the ProLiant or whatever. kind of touch to customers, which led us to have a standalone organization with a general manager and all integrated, specialized SEs so that we could touch the customer with experts and mimic Cray within a larger organization, if you will, be completely specialized. And we created the

Apollo line. We started, we moved from there, and then we acquired HDI. And then just after I left, we closed the acquisition of Cray. And if you look at all of this consolidated right now, it's basically essential geopolitically for America Inc. versus China. It's essential for some very key segments of the market to supercompute a lot of applications at the core and also with a lot of links to the edge. And it bridges beautifully with applications that went on the cloud. So it has become really at the center of the compute universe. And I think it's been accelerated over the last 10 years. Fascinating. And I'm sure Shaheen will have some comments in agreement with that. Absolutely. Yes. It's all music to my ear.

Exactly. Now living in Europe, it's interesting because it started in America with the SGI and the craze of the world. And then HP made all these acquisitions and became the powerhouse in HPC. But we saw as well China developing in parallel. And now it's a two-pony race between China and America, which is very interesting to watch. And all the supercomputers in the U.S. are production-centric. A lot of the ones in China, which on the paper may be more powerful, are more pilots, or they are not all in production. So I think the U.S. still has an edge.

But it's interesting to see what's happening right now in Europe. Europe feels they have a huge handicap for the future. They just signed a contract for the largest supercomputer in Europe, which is going to be made between two European vendors. And I think there will be much more to come. So Europe is trying to recreate the supercomputer industry that they had in the 80s. Japan has kept a foot in the game as well, and is very active. So it's a fascinating business because beyond what an individual developer or customer would need, there is this geopolitical dimension. And quantum, we take it to the next level eventually. But for now on silicon, it's the race that is geopolitically important for everyone.

So it is a business multifaceted with a lot of strategic dimensions. And so far, the U.S. continues to lead. The silicon is an essential ingredient. Obviously, the NVIDIA, the Intel, the EMDs are all American. And this is going to be very helpful. But we need to understand that this is a race and the race will keep going for the years ahead. Your comments about China's exascale systems versus the West's. I've heard that the Chinese

exascale systems are used really as specialized systems, not production systems. Is that what you were saying as well? That was, you know, we were frightened when we were comparing spec to spec, like, you know, the latest systems being released in China, we felt, whoa, they might be getting ahead of us. But when we were trying to collect information, what are they actually doing with them? We couldn't find anything really else. So they were really making a lot of effort to be able to put on the paper very exciting numbers. But whether they would be actually operating or not or capable capable of operating or not, was still to be proven. So I don't know.

Probably not for most of them. In general, as Alain mentioned, those are more projects than products. They look like they are one-offs. And of course, we right out of that analysis. Yeah, I agree. Now, our second question, Alain, would be, how has the evolution of supercomputing impacted industries, including scientific research, healthcare, finance, and manufacturing? And what potential do you see for further advancements in these sectors? Well, supercomputing has allowed to make calculations possible.

What we take months to weeks and days and hours. So obviously, it has allowed industry, pharmaceutical, intelligence, weather, all atmospheric based stuff, banking to basically become completely digital. So there has been a lot, a lot of progress. Things have been going very, very fast. And there is no limit to it else than right now, probably power consumption is becoming a big issue. But yeah, industries have really made a huge progress owing to the capabilities of computing. And this is why they have kept it in-house, because it's a core competency. Alain, you mentioned the importance of HPC and IT in general. And with that is a convergence, perhaps, between commercial IT and HPC, as was practiced before. I'm sure this was,

of course, part of the analysis that you indicated that had happened at HPE. This also includes the software stack and the convergence in that. Can you speak to how you guys looked at it and how, in fact, it panned out? Yeah, the software stack was obviously something we didn't have, to be very candid. So it was a headache, and it was the case to acquire Cray. Because Cray had a fully integrated stack that they had built over the years. HDI didn't have it as part of it. And I didn't have the resources to be able to build something like Cray fast enough economically.

So the Cray stack was probably the only one which was end-to-end complete for HPC. And their strategy was probably similar to the one initial strategy that NVIDIA had when they got into the GPU business, where they said, you know, we're not going to offer GPUs only, but we're going to offer an ecosystem that allows the customer to develop. I think Cray was doing something similar at the time with their own ecosystem. So that's how we acquired. I am not sure there are a lot of stacks that are equivalent commercially available. You may know better than I do, Shaheen, but I remember that Cray was really first in class on that in this dimension.

And it's essential to have a stack so that customers can develop really on top of it versus having to deal with the whole stack themselves. Definitely at the high end, they had a very concerted effort over many years. You're absolutely right. You know, the other thing you mentioned was just the general consolidation in the market. And HPE was the consolidator for a very long time. I remember joking that if anybody can design any systems, HPE is going to buy it and leave nothing else. And of course, over time, that in fact kind of happened. There is really no alternative at the very high end, except really a couple of few players, as you mentioned, a couple of Europeans, a couple of Japanese, and HPE basically is the main player, right?

But we were really focusing on the high end, because this is how the company was structured. It gave space to people like Supermicro, though, because HPC used to be for the big guys. But as it pervaded the industry, it became more of a meat market and then even a mass market today. And it's very hard to manage this at the level of a catalog. It means a lot, a lot, a lot of different configurations. And this is where Supermicro, for instance, managed to create what used to be a niche and which is now pretty big. So yes, HP and Dell in the first phase have totally consolidated the market. But today you see a third player emerging very fast, Supermicro, which has created

its niche where the other guys didn't play, the bottom of the market, and the cloud vendors, not very profitable for the big guys. And these are the two angles that have allowed Supermicro to basically become a significant player right now. So we will see how HP and Dell react over time. But in the meantime, Supermicro is having a lot of growth and allows customers to do a lot of different configurations with the Chinese menu of products, which is a match in terms of its diversity.

I think they've done a fantastic job. You know, in Dell, especially with the advent of generative AI and the explosion in demand for Gen AI and large language model workloads, Dell is making a lot more noise about being in the HPC business as it relates to Gen AI. They should. Dell was the best player for supply chain

and marketing-based compute. In a way, they started from being catalog online type of company. So their core knowledge was they were the best on the cataloging type of business, while HP would have been more on the added value side.

But right now, when they look at where the market is going, they have to become more flexible and to be able to offer this breadth of products in the high end, in the low end, on the catalog, but as well becoming more diversified in terms of platforms and infrastructure. And with generative AI, as you rightly say, this is going to be absolutely mandatory. So, you know, I am no expert in what is developing within Dell, but I would expect that if they want to keep their business developing, they absolutely have to go that way. They just announced a partnership with NVIDIA in GenAI. NVIDIA has been very smart playing with the so-called OEM,

Dell and HP at the time. There was a kind of a role model between Intel and the OEMs that was established by Intel and Microsoft for a very long time, for the former 30 years, since the 80s, right? There was this role model where the silicon would come from Intel

with a reference design, and then the OEM would build the platform on top of it, and then Microsoft and so on would do the software. Jensen was really smart because he never confronted the OEMs with that model directly. He never said, like, you know, I'm going to do full system. He did do full system, but he said, this is just for low customers to play, right?

In reality, you guys are the key players. But in parallel, he built complete applications, completely having known how to build an end-to-end business with the game business. He did the same in professional businesses, which then became AI-centric.

And he is now offering full solutions, as you know, fully integrated in competition with the OEMs. Now, it's very high price, whatever, but there is such a high demand that he can't supply. He could stop there and believe, I'm at the top of the hill.

I'm the most valuable silicon company. Well, silicon company. I stopped there. But when you look at the moves he made lately, I think he's using Supermicro, but he's not even looking at that at HP. He's looking at the cloud guy because he knows that AI is going to be offered on the cloud. And he is going for the battle of offering AI on the cloud.

However, the problem is that he also understands that as they did with CPU, the Googles of the planet are going to develop their own GPU. So he's not willing to invest on the relationship with these guys too seriously. Otherwise, he knows that will happen to him what happened to Dell and HP, who were before the arms dealers of the cloud guys, and then got replaced. So not to be replaced himself, I think his horizon is, I'm going to be the Google and the Amazon of AI on the cloud. And that's what he's looking at. Whether he can use Dell or HP or Supermicro to get there, but that's where he's trying to get. For him, competition is the cloud guy. Well, and they've invested heavily in CoreWeave, the GPU cloud company. And as somebody observed, I wonder how much of that money that NVIDIA invested in CoreWeave

will simply be used to buy NVIDIA GPUs. For now, yes. Because for now, that's the only game in town, right? So he's on top of the hill. But he is so smart being able to play with two or three horizons ahead. So you see the landscape has changed because you would look at it 10 years ago, Intel was the king. You look at it five years ago,

you would say the cloud guys are the kings. You look at it today, you would say NVIDIA is the king. Five years from now, who's going to be the king? NVIDIA will remain the king in five years if it is basically the default offering for AI on the cloud.

If he's only a GPU provider, what happens to him? What happens to Intel with CPUs? He will have his own AMD, which actually may be AMD, but it will never be able to keep 90% market share. The other guys are working so hard on it to capture a share away from him

and they will get there. And then longer term, you have to see what happens with Arm because they have a very different business model. But now that they have quasi-saturated the consumer side of the business, they need to go for the professional, the enterprise side of the business. So I think Arm is also a player and no question. Jensen wanted to buy Arm three years ago, two years ago, right? He could not for reasons beyond his control. But Arm is also going to be a player moving forward. And when you look at Arm and you look at Nvidia and so on, it means fragmentation of the market.

So while there has been a wave of consolidation on-prem and on the cloud. I think the future will be different and it will be towards fragmentation again. And the geopolitical crisis is going to even further foster the fragmentation because the Europeans need to have solutions. They cannot rely on Americans. I'm not saying they cannot. I'm saying they don't believe they can. The Chinese certainly are going for their own things.

The Japanese have still never shut down their own industry and continue. And then you look by segment, by market. So I think it's going to continue to be a fascinating industry until the time quantum happens as a reality or not. And that's the next frontier. What happens then? Is everything else becoming obsolete? Is the world we have known today obsolete and it's like a new beginning?

Or it's not going to happen? I don't know. I've seen them running, but not being used for operational work yet. But this may be happening. And the day it happens, it may be a new beginning for everything, for HPC, for AI.

It's interesting. A lot of money has been invested into it already. Alain, in our pre-call, we were just discussing what's happening these days. And you shared what you've been up to recently. And part of that conversation was also co-design and the idea of collaborations and co-creation and how that has been emerging in your view. Would you mind speaking to that again? Yeah, we come from a world of, I wouldn't say one size fits all, but I would say a world where you basically can look at the catalog and decide which product you like on a catalog. And then they can be, you know, bells and whistles that you can play with, but you are within the boundaries of a catalog product. The OEM does a platform and

they offer a few variances. This with the emergence of HPC, the future emergence of AI, and the fact that everything kind of stand out is on the cloud and everything kind of special the customer would like to keep in house make the business becoming much less of a catalog business but much more the custom and this is a scary world for most people so if we don't say custom we say deep chinese menu basically the capability to design something pretty tailored to a dedicated application, right? I believe that this is where the market is going on track. It has already started from where the emergence of Supermicro. the mass market, to enable many, many developers to configure agnostically whatever would be best

for them and see, OK, as a result, this is what I really need. How can I get it done? Versus this is what is on the catalog. Which one do I need? That's the shift, I think, that is taking place at the moment in the market. We are trying to build a company, to develop a company that is going to be the other side of it, which is we're going to do it with you. We're going to take this journey with you. We're going to co-design a system with you. You can start the journey online and you have a super configurator, but if you believe it's something bigger and you're willing to spend more time on it together with us, then we can put teams to work with you and find the best components on the market to put them together and go up to manufacturing and delivering to you wherever you are on the planet. It's a very interesting supply chain dimension. It's a very interesting supply chain dimension. It's a very interesting

front-end marketing situation. And it's in the bull's eye of where the market is going to, which is giving developers the capability to be very agile with no risk and to be able to actually get the hardware they need to fulfill their applications, not going through the limitations of cataloging. Alain, supercomputing often relies on massive amounts of data processing and storage. How have you seen data management and storage technologies evolve to support the demands of modern supercomputing applications? And what challenges are ahead? One challenge is going to be certainly on the energy consumption side.

You need to have layers of storage that cope with it. It's easier to compute fast than to retrieve data at the same pace. And so you need to move data to the cache and whatever. So a lot of work has happened here. But I think that the bigger the computers get and the storage farms attached to them,

the live storage paths attached to them get, the more you have a heating and power consumption issue. And that's going to be the next frontier on the hardware front. Right now, it's a bit of a zoo. You have all these liquid cooling, oil cooling, I mean, all these types of different solutions, but nothing is really market standard. It's probably a little frightening, but because this only relates to the very, very big supercomputers towards the exact scale, these guys are not scared because they can tailor a solution

to their system. But when this gets down to the CPUs, GPUs in the 500-watt range and beyond and whatever, when this becomes mass market, this issue is not going away. These things have to become intelligent and are going to help us managing that. There is room for stabilization of technology there. I think it's going to be very important. With SC23 coming up, what should we all look for in terms of areas where you think breakthroughs may be in process for the next generation of HPC?

Yeah, I think it's going to be interesting how HPC and AI start to blend together at the customer level and what kind of solutions emerge there. The HPC market has actually matured very well. So it's a market that is segmented with key providers. You know what you can do on the cloud. You then have specialized HPC cloud, if you wish. So it's been matured pretty much. What has not been matured yet, and you call it HPC or not,

I think yes, because it's high performance in whatever we look at in its compute. What has not been structured yet is how you go after AI if you are a customer. So I see 23, I'm a customer if I go there, right? So I'm going to try to find solutions. Who are the people who can help me to get on the AI wagon so that if my boss comes to me and say, you know, we need AI in the company, I have an answer. And it has to be a step approach. I think at the beginning, I need to find people who can help me imagine and kind of understand

what I'm doing right now and how AI can make it better. So it's kind of a phase of co-thinking, co-elaborating, you know, how could AI be helping my business? Once you understand it better, then the next question is, how can I build a solution? Who is going to be able to help me build a solution? And then it will probably come back to the initial thing. Then it's an HPC deal, right?

It's like I build a supercomputer to make it work with more GPUs and less CPUs, but it's kind of the same thing. But I think customers, except the very large and very specialized ones, are going to need a lot of professional help to be able to flush in their own business what it takes to leverage AI in a way that they're going to be much better faster. If I was a customer going there, that's why I would try to understand. And again, the company, I'm a board member of right now, SourceCode. We're just announcing the creation of three AI competency centers around the world because we think that customers need to have people they can talk to who know what they're talking about and can be anchors in their thinking

and be a sounding board in this first phase of what to build to make my company better in AI. They will probably find that there will be a lot of AI glue on every booth, right? It will be AI something. They should try to cut to that and understand which of these booths is having people who can talk to me and have an expertise that I can leverage in helping me to co-develop something that makes sense to get my AI business better. I think that's what's going to be hot this year.

But Shaheen, you are more engaged on SC23, preparing a few players there to participate. What do you think? I think you're absolutely right. I think the complexity of multiple complexities. One is just how do I go about AI? Some people already know because they're already leaders and they will be at SC23. But really, the bulk of the market is going to have to figure out how to do, as you mentioned, and how to approach the AI puzzle. And where am I going to get the best return immediately and show progress so I can do more? The second problem is, what do I pick? There are like hundreds of chips and hundreds of software packages and hundreds of how do I navigate through this maze? So to have somebody who is technologically deep and who can help you,

and maybe they have even kicked the tires and know what works and what doesn't work, I think is going to be really important. And then you have to find somebody who can get you an NVIDIA system to play with. Well, in fact, you know, Doug and I have discussed... If you know what you want to do, you can't, because you can't find the system. Yes, in fact, as you mentioned, Doug and I have talked in previous episodes that the shortage, the acute shortage of GPUs and accelerators in the market is really a strategic problem because it is causing customers to have to look at alternatives

when they might not have in the past. So if you have to wait 52 weeks to receive delivery, it really is a big incentive to say, okay, my God, maybe I do need another choice. So I think in fact, being too much in demand is not really so great for NVIDIA. I think being just enough in demand to be able to fulfill it is good. But now I think it's opening the door for competition. With that really red hot demand for NVIDIA and for GPUs, it's certainly, I mean, the other Intel and AMD, if we're looking for a change that's going to come about in the next two or three years, that certainly is more GPUs, more GPU alternatives on the market.

I think we can certainly expect. This will happen. Yeah, no question. A hundred percent and very quickly. The problem is the ecosystem, the stack, the software stack. This is going to be longer. It's probably going to be next AMD

and next Intel, but AMD is not the software house, right? So it's challenging their skill set to get there quicker. Intel probably has got more capabilities at the software level than they do, right? So it would be interesting to see who is really the next contender, the next follower. But, you know, right now the lead times are very, very, very, very long. And as Shaheen says, this is not good for NVIDIA either.

And, I mean, NVIDIA is missing a lot of business. They have much, much more orders than they can satisfy. But imagine what it could be if there is more tension in the Taiwan Strait. They will go there while Intel is building its own path. So I'm not burying Intel yet for the long term with all the investments they are doing. So they may not be the innovator and the first mover any longer, but once the market is established,

they're going to be able to provide the cavalry, right? And that's why NVIDIA has got to get to another place. If they become cloud-centric, that's probably their real destination because in the silicon level, not doing their own silicon with Intel, who will be doing their own silicon and finally end up catching up

after two or three generations lost they have got? Over time, I think the fabless companies with the fabs being outside of the country, we have some forms of risk and challenges that the customers will need to buffer. And so they will need to mitigate these risks. And Intel will remain a very strong company in the business. So there will be room for everybody to play.

It's not going to be one. I mean, at the moment, it's like Tesla for electric cars, Nvidia for GPUs, whatever. But the world will keep moving. There is always action. We will never get bored in this industry. That's the most exciting industry on the planet and it will stay that way. By the way, on sustainability and on cooling and the problems

of cooling as these systems get denser and bigger, how about a data center under the Arctic ice cap? That's already been done. Or in the ocean. Or in the ocean, yeah. Then been done in the ocean. Oh, in the ocean. Oh, in the ocean. Yeah. Oh, in the ocean. Yeah. Yeah. Yeah. Then you have a latency issue.

Microsoft, actually. Microsoft, I don't know where. I've been doing a lot of work with different kinds of data centers, but there is no, I mean, maybe I'm not current here and I don't know the latest, but I have not seen anyone becoming the de facto standard in fighting this issue. It's case by case, as Shane was saying, data center by data center. And each time you go through a long list of solutions and you can pick the one you like. It will be different.

But right now, that's where the market is. The edge will be in standardization when somebody succeeds to be the de facto play then then this player will have the okay and then i had one other question and maybe we're going on too long but um with ala having been in and around HPC for so long, I'm just curious your reflections on this explosion in Gen AI, ChatGPT, large language models, and whether you think, Alain, this will have the impact that some say it will, that it's really a new chapter in IT.

I have read two or three books lately on the question that you just asked, Doug. One extreme thinking is that this development is going to enable the next evolution of the human race, right? So the level of expectation is insanely high because for the first time owing to Shad GPT, most people have understood what AI means. Until now, people had no idea. But then, you know, when they started typing a question or putting a page, you know, and then asking for the corrections of the autograph or whatever, people were like, it means that computers are going to be smarter than us. Really? And so from that dimension, I have read a lot, and it's, okay, you know, who knows?

Who knows, right? I mean, the only, and the conclusion of everything I've read on the subject is that, well, computers are going to be able to do everything we have, human beings, else than conscience. Conscience won't be replicated. So faith, conscience, love are going to be the most difficult things to replicate.

Okay, so this dimension was the one that had most interest for me, actually. Then if you look at how it will impact IT, I tend to believe that it's going to be like another wave. You know, it starts with a few initially who are going to build solutions around that, and then it will slowly go to the tail end of customers. So it's going to take a while, but it's going to take a lot of education for people to understand how they can truly leverage that with a business outcome and make money out of it.

Outside of the usual suspects, we'll be able to leverage most of it. So I think the excitement was probably more for the philosophical reasons that created this huge boost initially and then for business reasons. The business reasons are going to come, but it's going to take a bit longer and it's going to be a tail and it will impact IT like everything else. Everything ends up being on computers at the end of the day and it's going to make the demand for the business bigger and bigger. Look, have you seen the market growth data? Like you look at IT and you look CPU-based, GPU-based. You see CPU-based, I'm simplifying. For the last three years, it's flat. It went down a little bit with COVID, and then it came back up

kind of flattish, right? It's GDP growth, GDP growth. Then you look GPU-based, which would be AI-like application. And it's like on top of it, you add a new market as big as the other one in five years. Yeah. Huge. That's what market analysts who know much better than I do see right now. That's what PE firms, VCs, and whatever do. This is why they're banking a lot of money on AI, right?

There is no reason to believe that this is not going to be the case. But it's not going to kill legacy infrastructure type of IT that has digitized the business, digitized the world, right? This is going to stay and grow at gdp speed but this is like a new layer on top of it that is going to be probably as big as the layer we have right now because intelligence is going to be embedded in so many so many things but ultimately the destination is what fascinates me i mean is this going to replace us? How many people are still going to have a job 20 years from now if AI delivers its promise? Who cannot be replaced by the machine? Right.

We just did a video on the legal profession. It's already having a big, big impact on how long it takes for lawyers to complete tasks. And it's impacting their billable hours and all that. But, you know, it's funny or interesting. You began your remarks talking about that HPC has emerged from a niche industry to really much more broader base. And that's just been an ongoing process. But certainly the Gen AI boosted HPC far more into the mainstream than anyone else that's come along.

It always was. But yeah, you're right. If you brand generative AI HPC, then it is most of the market today. Legacy HPC plus generative AI is becoming most of the market. The rest is just legacy apps, right? I think IDC just put out a report saying Gen AI is growing with over 73% compound annual growth rate. Wow. But, you know, I think we need to come back to the existential questions. I mean, you know, there are a lot of people voicing opinions, whatever, including Mr. Musk, you know, saying we have to be careful, whatever.

I think this debate is right. There is a valid debate here. But we have not even noticed that IT has got to the point of making the world already unbearable to a lot of people. My mother is 89 years old. Do you think she can go on an iPhone anymore? She can dial a phone number, but can she go into all these windows?

No. So imagine when us guys, you're younger than me, we'll be 89 years old. Is AI going to help us to be alive and give us basically, you know, I am hungry. I want to pay my taxes. What time is it?

This kind of thing. Is it going to help us being better humans or being humans longer? Or is it going to make us completely irrelevant, completely obsolete, and we are like subspecies? These are all questions that this industry is leading us to think about. Because when you look at where we were, you know, when I started in this business in the 80s, where we are, when is it, 40 years later, where we are today. You know, I started at Texas Instruments and we were doing learning aids and calculators at the time. And we were thinking, where is this going? And we were always talking about one thing, which was the ultimate portable electronic object,

which translated in today's world is what Elon Musk does with Neuralink. You basically don't have a calculator. You don't have an electronic watch. You don't have a phone. You have another device, which is in you you know it's a bracelet or it's in your body like my dog she has a tag right you know it can be something like that so you may have your tag and your this tag may make you a superhuman and so we become an edge device i am an edge device i'm 89 years old as an edge device. I'm 89 years old as an edge device, right? I have my small computer. I decide what I compute in my own brain with my small device, helping my old brain, compensating my Alzheimer and making it work. And I decide what goes in the central computer of the nursing home, right? While the central computer of the nursing home send me links like, Alain, time to take your pills.

Alain, time to go to the dentist. So that's what's ahead of us. So HPC is going to become our life. It's going to enable our life in the future. We're going to be HPC enabledenabled humans, the next generation of human beings. Well, it already is. Something like that. That's exactly right.

It's going to become everyone's life, exactly. Speaking of Edge, Alain, your views on the place of edge vis-a-vis the core data center, the cloud, how do you see that evolving? And what are some of the key issues there? First phase was, you know, let's connect the devices to the central farm. Makes sense, right? So billions of devices connected and all sending data to the ranch and the ranch sending orders to the devices, phase one. But as there are more and more sophisticated devices, you start to scratch your head and wonder whether it makes sense to send so many data to the ranch. Or do you want the ranch to know so many things?

Or do you want to have so much traffic on the network to go to the ranch? Or do you want the central computer to heat up so much because there's so many data to compile? Or would it be more efficient to do the work on the device? And as of today, this is the ultimate device, right? The iPhone or whatever work on the device. And that's, as of today, the ultimate device, right? The iPhone or whatever is the ultimate device. Or the car, the electric car is becoming the ultimate device as well. So I think that the current trend is that there is much more computing, data, data management, data, you have a word for it, maybe, Shahin? In situ. In situ, data management, data, you have a word for it, maybe, Shahin?

In situ. In situ, data management. This is where I think you see a bifurcation between the devices. There are devices that are just a connection, so that this analog world connects to the digital world. And you have devices that, beyond being a connection, are also a computer on their own. And this computer becomes an HPC one because it computes more and more and more. So soon a car will be a supercomputer.

Maybe it is today. It depends which definition you take. So I think that's kind of what's happening at the edge here. Interestingly, it's a very tough business. If I take the vendor path, it's a very tough business because I remember at HP when, you know, and at this segment of the market, we had a fantastic core product offering. And Dell had a very good one. I'm not saying HP is better. I'm saying we both had a very good engine, if you will, to put in a device. But the problem was to find applications that would be generic enough that the customer would be committed to a volume and you could basically do a business out of it

versus spending so much money in pre-sales educating the customer and the customer basically in reality doing a proof of concept versus really doing something that that works right so this business has been a headache because everybody said iot is the next frontier everything blah blah blah in terms of business it has not turned that way so far very fragmented nobody is going to be able to, you know. My view is similar to where the central farm is going. I think it's going to be a co-design issue as well. Each device is going to need something a little bit different. And very important for a systems vendor to have the capability to have a ruggedized small factor solution that allows a small HPC farm at the level of the edge device.

But there the constraints are there. You cannot have hit things or whatever. Right. So it's the body that hits up and that calls the device. And in a way, it's different, different form factor or smaller. It's different constraints. But it's fascinating also what can be achieved at the edge and what it belongs in the making. It's taken a long time to take off, but it's going to be a big part of the business moving forward. And this will never go on the cloud.

We talk to the cloud, but it will be a second life for the systems vendor if they need one in case everything ends up on the cloud, which I don't think it will. And then you could ask, you know, how is AI going to impact the edge? Well, again, remember me at 89 years old with my device here, right? I would be an edge device in the world of AI. So it's infinity what AI is going to allow us to cascade to a new paradigm,

but the architecture remains the same. We have this triangle between the cloud, the core data center or the data centers, and the edge. These are the three dimensions of the triangle. And you have to be in the middle. You have to have your brain in the middle and sense where you do what and why, what is more economical, what is safer,

what is more logical. But this is a triangle IT people have to deal with before or after the chat GPT revolution. Fascinating comments. That's excellent. Yeah, absolutely. In part because I agree with almost all of it. Okay, thanks so much.

We've been with Alain Andrioli. Great talk. Thanks so much. We've been with Alain Andrioli. Great talk. Thanks so much. Thank you, Alain. It was fabulous. Thank you, guys. It was a pleasure. Bye-bye. That's it for this episode of the At HPC podcast. Every episode is featured on InsideHPC.com and posted

on OrionX.net. Use the comment section or tweet us with any questions or to propose topics of discussion. If you like the show, rate and review it on Apple Podcasts or wherever you listen. The At HPC Podcast is a production of OrionX

in association with Inside HPC. Thank you for listening.