Exascale to AI: Inside HPE’s Next Era of Supercomputing - Six Five On The Road at SC25

The future of scientific computing now hinges on the intersection of AI acceleration, exascale architectures, and the infrastructure innovations binding them together.

‍

From SC25, host David Nicholson is joined by Hewlett Packard Enterprise's Trish Damkroger, SVP and GM of HPC & AI, for a conversation on HPE’s leadership in exascale computing and the technologies powering their innovative scientific breakthroughs. They highlight HPE’s key advancements, including the Cray GX5000 systems, Slingshot 400 interconnect, next-gen DAOS storage, and unified management software—all engineered to address customer challenges around density, cooling, energy efficiency, and secure research collaboration in the age of AI-driven science.

‍

Key Takeaways Include:

‍

🔹Continued Supercomputing Leadership: HPE maintains the record for the world’s fastest supercomputer, showcasing continued innovation with recent projects like El Capitan, Frontier, and Aurora.

🔹Evolving Customer Demands: New challenges are emerging as HPC and AI converge, with customers seeking higher density, advanced cooling, energy efficiency, and streamlined, secure global collaboration for research teams.

🔹Next-Generation Solutions: HPE’s Cray GX5000 systems, Slingshot 400 interconnect, and DAOS-based storage deliver enhanced performance, manageability, and reliability designed to meet the requirements of AI-powered, high-density environments.

‍

Learn more at HPE.

‍

Watch the full video at sixfivemedia.com, and be sure to subscribe to our YouTube channel so you never miss an episode.

‍

Or listen to the audio here:

‍

‍

Disclaimer: Six Five On The Road is for information and entertainment purposes only. Over the course of this webcast, we may talk about companies that are publicly traded, and we may even reference that fact and their equity share price, but please do not take anything that we say as a recommendation about what you should do with your investment dollars. We are not investment advisors, and we ask that you do not treat us as such.

‍

David Nicholson:

Welcome to Six Five On The Road. We are here at SC25, the supercomputing conference in St. Louis this year. A little bit of SC trivia. St. Louis is actually the home for the supercomputing organization behind SC25, and I'm joined by a very, very special guest, Wonderful to see you again. Trish Damkroger, All Things HPC at HPE. Welcome.

‍

Trish Damkroger:

Thank you.

‍

David Nicholson:

What have we got going on this year? SC25? What's up in the El Capitan world?

‍

Trish Damkroger:

Well, what's up? It's at the top again. I mean, that's great to have El Capitan be our number one on the top 500 again. We also have Frontier as number two and Aurora as number three. So again, we have the top three exascale machines in the world.

‍

David Nicholson:

And I couldn't help noticing with all of the talk of 10 gigawatt data centers that El Capitan isn't very thirsty, really. It's only 30 megawatts or so.

‍

Trish Damkroger:

Right.

‍

David Nicholson:

Only?

‍

Trish Damkroger:

Only. You know, again, it's a small city, but yes. And power is going to be driving the future for sure, what we can really fit. And in fact, we had a great conversation last night over dinner about the sensational demand for power and how are we going to get there, especially in Europe and some of the other countries that are pretty tapped out.

‍

David Nicholson:

Any thoughts on how we're going to get there?

‍

Trish Damkroger:

Well, I don't know. I mean, I was at Davos a year and a half ago and I came home from Davos and said to my husband, we need to invest in small nuclear reactors. That's the future. And so we'll see. We'll see if nuclear, if we're going to be able to do it by other means. I just really hope we do it in a sustainable way. And we are working with some companies about that. Do heat reuse effectively with some of our Mod Pods right now? So we are.

‍

David Nicholson:

Yeah.

‍

Trish Damkroger:

We provide data center facilities like UK Bristol where we actually stand up the building in their parking lot and then we put the compute into that building and you do heat reuse with our amazing partner Danfoss. So we have a whole sustainability story that I think is going to be even more important as we move forward with this growth in power.

‍

David Nicholson:

So people are familiar with HPE of course. HP, HPE, long history there. Some people might forget about this, the Cray relationship and how you've got the Cray DNA that's a part of this. How much of that has sort of found its way into El Capitan as it exists today?

‍

Trish Damkroger:

Great question, because I think, I mean, number one, this is 50 years of Cray. Cray-1 was developed in 1975 and actually deployed at Los Alamos National Lab in 76. So we have 50 years of this technology. And if you go to the Cray Museum in Chippewa Falls, Wisconsin, which is where our manufacturing site is, you'll see that liquid cooling was part of that way back when. So 50 years of liquid cooling technology that we've continued to develop upon because dense compute is what is needed for both modeling and simulation and these AI workloads. So I would say that the trajectory of liquid cooling is definitely part of El Capitan. The dense compute, tightly coupled, knowing how, you know, just the fabric and getting everything to work together. The Cray-1 is in a circle, so you didn't have to have the long cables. And so that continued philosophy of close coupling is part of El Capitan also.

‍

David Nicholson:

We can't imagine that anyone was really behind branding at that point in time when the Cray-1 came out. You think of a bunch of engineers not caring about that. But when you hear the word Cray, I think we all think of that physical design that's so unique.

‍

Trish Damkroger:

Right, with the seats.

‍

David Nicholson:

Yes, sort of emblematic of what a supercomputer was meant to be. It would be interesting to compare the horsepower of that system with what we carry in our pockets today.

‍

Trish Damkroger:

Oh, well, trust me, yes. we're a little bit more powerful in our pockets. But if you like that design, you'll have to go to our booth because of our new GX system, GX 5000, which I will be proper and say the HPE Cray supercomputing GX 5000, which is going to be our next iteration for our dense platforms. We have kind of modeled in the middle of our booth this year that's along the lines of the Cray-1. So there's actual seats.

‍

David Nicholson:

Oh, OK. We'll have to check that out. Well, on the subject of liquid cooling, is it pretty much a given that in order to keep up with the sort of densities and performance requirements that we want to have over the next several years, that five years from now, pretty much everything's going to be liquid cooled in data centers?

‍

Trish Damkroger:

You definitely want to be a part of AI if you're going to be part of GPUs, but even our CPUs are becoming, our thermal density of the CPUs are driving to liquid cooling. So yes, I would say that we're going to see, we have seen a growth in liquid cooling, the demand for liquid cooling. I think the biggest challenges are getting data centers that are capable of liquid cooling. And we work with them. So we can actually do that through our rear door heat exchangers and our whole arc design that can help data centers do liquid cooling within an air cooled data center. But yes, you're going to have to have liquid cooling in order to be able to dissipate the heat just generated by these dense compute platforms.

‍

David Nicholson:

And what you described is, you could almost apply the term hybrid to liquid cooling in the sense that you offer the chance to obviously cool at the chip level, at the cabinet level. Exactly. all about dissipating that heat from wherever it exists. In the supercomputing, high-performance computing area, traditionally, that's been a CPU world. A lot of the buzz now, when we think of anything high-performance, we immediately think, oh, it's got to be GPU, it's got to be AI. But we're here at Supercompute, where some of the national labs come in and show off what they're doing, top academic institutions are showing off what they're doing, I'll copy time. What does that transition look like in the HPC world, in terms of the mix of CPUs and GPUs? Are CPUs going to be around for a while?

‍

Trish Damkroger:

So we still do have customers that buy almost 100% CPU systems, and it's really dependent upon the applications. So if they have homegrown codes that are written for CPUs, and they have not ported those codes to a GPU, they're going to stay with the CPUs. Weather is one of those. We have a lot of our defense clients that still have a very strong CPU, so completely dependent upon the workload. A lot of the national labs and our more general user science base has been pivoting more and more towards GPUs. And they are seeing the acceleration of GPUs. And there's a lot of demonstrations of a huge speed up by being able to take advantage of the parallelization of a GPU. But again, it's a lot of work to get there.

‍

David Nicholson:

So essentially, you're talking about rebuilding the code that's accessing that parallelism in order to get those advances. Is it generally something that can be done?

‍

Trish Damkroger:

Yeah, it takes manpower, and I think that also we've been talking about the convergence of HPC and AI, and I like to say model and simulation and AI, because those are all HPC workloads. So we've been talking about this convergence for a couple of years, but we're really seeing that now. We're seeing where people are using AI to develop the data that they're putting into the modeling simulation and vice versa. We're using it to steer modeling and simulation. We're accelerating the science outcomes by bringing in some AI techniques. So there's a ton of this convergence going on, and that's where GPUs are very useful.

‍

David Nicholson:

Last year, there was a bit of a buzz among some of the attendees that there was a fear that all of the discussion of AI and GPUs was kind of crowding out the traditional HPC workload conversation. There were concerns that folks would be chasing the shiny object in terms of AI and that they would be neglected. Are your customers that are in that space, are they feeling more comfortable that they're going to get what they need? Specifically, you would hear talk about floating point processes, 32 bit, 64 bit, fine levels of granularity that you need for critical research and modeling that isn't needed for AI. So have we struck a balance there in terms of supporting those folks into the future?

‍

Trish Damkroger:

I think we're still seeing it play out because if you look at the market and what's driving the silicon vendors is definitely this AI. But It's not that the HPC or the traditional HPC has slowed down, it's just AI has added to it. When I was talking with some of the analysts just yesterday, they were saying that traditional modeling simulation has doubled in size from about 7% to 15% CAGR. So that is a growing market, that floating point 64 is a growing market. It's just the AI market is growing so much faster. So there's going to be a continual push play over who, you know, how much floating point 64 is going to be, how much of the silicon is going to be devoted for FP64.

‍

David Nicholson:

Yeah, well, so let's get back to like what HPE is really showing here at SC25. What are the headlines?

‍

Trish Damkroger:

The headlines are definitely that we continue to be the leader in supercomputing, and we're very, very proud of that. And that is the heritage of Cray, SGI, and HP all brought together. The other thing is we did announce our GX5000, like I said earlier, which is going to be our dense platform. It's 23% denser. than the current EX platform. So this is going to help you, and it also has warm water cooling, which is also part of this whole sustainability. And a lot of times that's needed in the European market and other markets. The other thing we announced is our K5000, and our HPE Cray supercomputing K5000, which is exciting for me because I've been part of this journey. It's our new storage platform based on DAOS, which is an open source platform. We acquired the Deos team from Intel a year ago, about a year ago, a little maybe like 11 months ago, right? And we've integrated them into our HBC storage team and now have built basically the Deos as a product. It's currently in Aurora. computer, the Exascale computer. And so we've had some great demonstrations and it's also at LRZ. So we had some POCs and a large platform that we've been able to work with Argonne and actually productize it and make it available to others.

‍

David Nicholson:

Do you realize that El Capitan manages to somehow get 11 million processor cores to work effectively together? I think you probably knew that.

‍

Trish Damkroger:

Well, that's what we do. We do big systems.

‍

David Nicholson:

Isn't that a lot? Isn't that a big number?

‍

Trish Damkroger:

It is a lot. And the thing that is amazing is, you know, it's proven in our LINPACK run for HPL for the whole top 500. We have to have all of those cores working together to get that number. So not only do we know how to deploy big systems, we know how to make them scale and work across the full system.

‍

David Nicholson:

So that's got to spill over into all things AI, because everything we hear about now is just about crazy scale. It's almost one-upsmanship. It's like, yeah, my neighbor's probably building a 10-gigawatt data center. That doesn't impress me. Only a 100-gigawatt data center would impress me. El Capitan? 30 megawatts, which doesn't sound like a lot compared to a 10 gigawatt data center, but that's but it's huge. It's massive. It's massive. And I guess the point is you have real world experience. So not just not just liquid cooling going back 10 years before I was born. If I remember the 50 years.

‍

Trish Damkroger:

Yeah.

‍

David Nicholson:

Yeah. Wink wink. But also the fact that you've been doing these things at scale for a very long time.

‍

Trish Damkroger:

50 years.

‍

David Nicholson:

Because last year, guess who number one was?

‍

Trish Damkroger:

Right, exactly.

‍

David Nicholson:

It's been a while. This is not new for HPE, which bodes well for the future in all things AI.

‍

None:

100%.

‍

Trish Damkroger:

And in fact, that is again, one of our secret weapons is our service organization, Deployment and Services. Because we know how to stand up these systems quickly. We do it in weeks. I mean, El Capitan, you're talking about that massive system, right? We had it stood up in mere months, and scaled, running the top 500. So working across the full system, and that's a lot of tuning for that performance. So we also do that with our AI customers that are buying these very large MVL72 systems with us. We can deploy them and get them working quickly for you, because it's time for science.

‍

David Nicholson:

Yeah, time for science. I absolutely love it. So Trish, the first time that we had a chance to meet, actually it was at HPE Discover this past year. When I heard about your background, I immediately just thought of referring to you as triple threat Trish from now on. And I don't mean to put you on the spot, but I think it's really important from a credibility perspective, because you're not just speaking from the perspective of the vendor HPE that has partners and customers in the world, but you've actually been there and done that. So can you kind of quickly go through those three aspects?

‍

Trish Damkroger:

Okay, yes. No, I'd love to. I've had an amazing career and I am very fortunate that I've been able to have that. So I started at the Department of Energy National Labs designing nuclear weapons. So that was where I started. And then using computation as part of that. Ending up running the computation directorate at Lawrence Livermore National Labs where we used high-performance computing as just fundamental for us in certification of the nuclear weapons stockpile. And I had an opportunity to go to be the general manager for HPC at Intel. And it was a fabulous opportunity because I really got to understand the silicon behind how you do dense compute. And so I spent five years there learning all those things. I get to bring all that experience together as the system provider at HPE, which I think allows you to really understand from the customer point of view, down to the hard technology as part of the system. And I think even within the organization I manage today, we rotate people from engineering into deployment and services so they know what the customers are going through. That's how we make better products and better services.

‍

David Nicholson:

Excellent. Thanks for sharing that with us.

‍

Trish Damkroger:

Yes.

‍

David Nicholson:

For Six Five Media On The Road, I'm Dave Nicholson here with triple threat Trish Damkroger. All things HPC, high performance computing at Hewlett Packard Enterprise. Thanks for joining us. Stay tuned for more.

‍