>> Announcer: Live, from Las Vegas, it's the Cube, covering AWS Re:invent 2017 presented by AWS, Intel, and our ecosystem of partners. >> Welcome back everyone, we're here live in Las Vegas for 45,000 tech industry folks and customers with Amazon re:Invent 2017. This is the Cube's exclusive coverage, I'm John Furrier, with my co-host Justin Warren this segment. Our next guest, Bill Magro, is the Chief Technologist for Intel covering HPC high performance computing. Bill, welcome to the Cube. >> Thank you. >> John: Thanks for coming on. You guys, your booth's behind us, I don't if they can see it in the wide shot, but Intel is really taking advantage of the I don't want to say Intel inside the Cloud 'cause that's really what you guys are doing, but you got so much compute, this is your wheelhouse. Compute is what Intel is. >> Bill: Right. >> Andy Jassy at AWS, talking with their customers, they want more compute, edge of the network, so HPC, high performance computing's been around for awhile. What's the state of the art and how should people think about HPC versus the Cloud, are they the same, what's the relationship? >> Intel actually thinks of HPC or high performance computing more in terms of the activity and the workloads than the infrastructure that it runs on. So very early in the days of Cloud computing, there were a lot of people who said that the Cloud was kind of the opposite of HPC and therefore, they could never go together. But we think of Cloud as a delivery vehicle, a way to get access to compute storage networking and HPC is what you're doing. And so then, if you think about HPC as kind of a range of workloads, you can start to think about which ones are a good fit for the Cloud, and which ones aren't. So we talk a little bit about the high performance computing and tailored infrastructure for the most extreme cases of HPC. That's where you see the biggest differences with Cloud, 'cause they're at opposite ends of the spectrum. >> But you see holistically the Cloud is interplaying with HPC. >> Yeah. >> They're not mutually exclusive. >> Absolutely, we see Cloud as a way to deliver HPC capabilities. So if you think of the most demanding HPC problems, the ones that are used in national security, that are used to design commercial airplanes, and so on, those are some of the hardest problems. Predicting the climate change, predicting the weather, paths of hurricanes, those are what we call grand challenge problems. Those are not running in the Cloud. Those are running on dedicated, tailored, infrastructure built for high performance computing at that extreme. And those systems have a lot of characteristics such as very high performance networks, different from ethernet, custom topologies and are designed with software to really minimize variation because it's one large problem that has to move forward. The Cloud is kinda the opposite in a sense. It started as taking a large amount of resources and making it possible to carve them up, right. It's the opposite of aggregating resources. And so that's where a lot of the early thoughts of Cloud and HPC being at odds with each other. >> It seems to be a dream scenario because I mean, in the old days, in 80s and 90s when I was breaking into the business. If you were a database guy or a compute guy, you were a specialist, it was high end kind of computing. Moore's Law, certainly Intel, you guys took advantage of it. But now, you see so much, it's cool to do more compute. So like, it's been democratized. Databases and compute, certainly in all the conversations, for everybody, not just, the technologists. >> Right, where that's where Cloud fits in for HPC. So if you think of HPC in terms of the characteristics of the workload, it's something that's really demanding computationally. The product of the computation is like an intellectual insight. You can design a better airplane wing, a safer car, you can figure out where that hurricane is going and tell which people to evacuate. There's an intellectual product to the compute. And then the last characteristic is when you apply more compute power appropriately, you get a more valuable result. So it could be better prediction of that hurricane path, it could be a safer car because you have more time, you have more capability and were able to build a better design ahead of that deadline to get that model year of the car out. And so, if you think about that, there's a lot, there's a broad spectrum and I talked about some of those most extreme problems, but even in something like designing an airplane, there might be 16, 20, a hundred different small design variations you want to explore. Well those can actually be great for The Cloud, 'cause they're small calculations and you run many of them at the same time. And the elastic capability of The Cloud augments the supercomputer that you might be using to run your hardest problems. >> So the aperture of problem-solving is huge now. >> Bill: That's right. >> You can do more. I mean we had Thorn on yesterday. Thorn was a company that partners with Intel to do, you know, find missing and exploited children. AI for good, so everything's possible. >> Yeah even AI we think of as an example of a high performance computing workload because what does it do? It gives you insights that you didn't have otherwise, it's compute intensive, and it does better when you apply more resources. So that fits our definition. So AI is definitely under the umbrella of high performance computing. >> One of the things, one of the great benefits of Cloud is the elasticity which you mentioned before. It's like, and some of the, we know that Amazon's just brought out the C5 Instances which is a specific instance site, which would be quite useful for HPC. But what is it about the bursting workloads or that elasticity that specifically works well for HPC do you think? >> Well, there's a couple use cases that we think are particularly relevant. One of them is an existing company. Just imagine some Fortune 50 manufacturer. They have a lot of stuff that they really need to build their own supercomputer for, their own high performance computing system, but their usage, even though they keep that system busy all the time, there is some variability and they have opportunity costs of an engineer sitting while their job is in the queue, 'cause you're paying that engineer but you're not giving them insights, right. And so the Cloud can augment that, but we have a lot of examples of large Fortune 500, Fortune 50 companies augmenting their on-premise with Cloud as a way to push those workloads that can run on the Cloud there, to free up those on-prem resources which are much more tailored, much more expensive and get more value out of them. >> Okay, and what's Intel doing to help customers figure out which of those workloads is best suited for Cloud and which ones are better suited for something which is running on site? >> Well, it's mostly through our influencer sales force who engages with many, many major companies and provides consulting, because Intel doesn't sell computers directly to anyone, so it's more of a knowledge, our knowledge and sharing that with people. And what we're trying to help enterprises understand is what workloads need to stay on premise, which ones can go to the Cloud and how this, the elasticity of the Cloud can augment those on-premise resources and thus, you know, go back and forth. >> It's the classic mission for Intel, make the apps go faster, faster, smaller, cheaper, right. >> And get 'em land in the right place. So really, the two biggest considerations we find in deciding whether a workload goes into the Cloud or stays on-premise in high performance computing are the following, one, is really the sensitivity of the IP. There's a lot of workloads that could run in the Cloud and people simply want to keep it on-premise 'cause they're more comfortable knowing that their IP is sitting inside their own firewall. Though the reality is, more and more companies are getting comfortable with Cloud security as they see data breaches. And realize that some of the big Cloud providers, like Amazon, maybe have better access to the security talent than they do. >> I think Goldman-Sachs just announced they're going all in. That's Goldman-Sachs, they never do a testimonial. >> So the privacy and the sensitivity of the data is king, you know, you have to be willing to put it in the Cloud. Then the second question is, is it a technical fit? And that's where this spectrum of workloads comes in. The bigger a workload goes and the more you want to speed it up but keep the workload the same size, that's what we call strong scaling and that starts to stress the network, and stress the system. And that's where these tailored systems come in. And so, you have to look at where things fall on the spectrum. A good example of workloads that would fit is these design space explorations, anything we would call pleasingly parallel or embarrassingly parallel in the industry where the communication does happen, but it's not the limiter of the calculation. So screening for a drug candidates, for personalized medicine, lot of life sciences applications, financial services is a good fit, in manufacturing a design space exploration maybe for different designs and materials for a dashboard or a component of a car. >> Bill, when you were at your Thanksgiving dinner and your family or wherever, you're moving around in your personal life, you're a technologist. How do explain the phenomenon of Amazon Web Services and the Cloud action right now? Because, you know, you're in it every day. You're close to all the action. But I get asked all the time, what's the hub-bub about AWS and it's hard to explain the phenomenon. How would describe the, I mean you're talking about tailored systems, elasticity, I mean it's a tech dream. I mean, how do you explain it to like a normal person? >> The conversation's usually pretty short because my family involves a historian, an English major, an accountant and people who really couldn't have, a musician, a singer, people who really don't have the slightest interest in technology. >> It's hard to talk about lambda, when you're. >> So I'm really the only technologist in my family so I just avoid it, but the question does come up with my parents. You know, parents like to brag on their kids so they like to know what you do, and every year my mom asks me what I do and I try to explain high performance computing to her and she says, oh, I don't get it. But when you explain it in terms of things like climate modeling and being able to support the nuclear test ban that's worldwide, that's done with high performance computing. Safing cars, finding missing children, better quality of life through all the AI that we're now experiencing. >> John: Analytics is a great use case. >> Then people say, oh, you know, they can understand the use cases. The elasticity of the Cloud, really is not something that I discuss with family, but even coworkers, I think, that's what the conversation focuses on. Recognizing that high performance computing is a range of workloads. >> Okay, so I'll rephrase it differently. What's your perspective on, what observations that you get excited about that are enabled now by these new use cases? 'Cause there's new things now that are possible. The number of computations, you got analytics, you mentioned a few of them. What jumps out at you, wow, that's really awesome, we can do that now? >> You know, this is gonna sound a little odd, and maybe not what you expected, but I'm not actually a technology enthusiast, believe it or not, despite. I think technology's cool, I like what it does, but I don't get super excited about technology. One of the things that I'm excited about with the Cloud is probably at the opposite extreme of what you would expect which is, back to, how does the elasticity of the Cloud fit? There's so many companies in this world who could benefit from high performance computing and don't today. A recent study showed that 95 percent of U.S. small medium manufacturers which is over 300,000 are not using HPC today. And so, as they're part of this supply chain, whether it's into a Boeing or and Airbus or a Lockheed Martin or a Honda or a Toyota, there's this whole supply chain. HPC's being used at the top, it's not being used at the bottom, so I think the Cloud is actually really, really exciting because it allows somebody to get over those initial hurdles, the cap-ex, getting access to pay as you go, prove the value proposition, because a small medium business actually has to take a risk to use HPC. They have to divert capital and divert resources. And they could lose a contract. >> So do you see a lot more companies starting to take advantage of some of this high performance computing capability just because it's now, you can rent it by the hour and try it out, give it a bit of a whirl, and then see, actually this is going to be really valuable for us, and then deploy a lot more of it. >> Exactly and that's one of the key things we're promoting is 'cause we want to bring more people into the world of high performance computing. So, AWS provides all the building blocks. Compute, elastic storage and so on. But high performance computing applications really expect a specific type of platform that they can run on, and that platform aggregates the resources so there's a number of companies Rescale is one, Psycho-Computing, and others who are actually providing that platform layer. And then once you've got the platform layer, all the, I'll call it the, geeky stuff that they do, AWS has abstracted away. Now the applications can run and that's that's what's bringing new users in. >> Bill, final question for you. AWS launched its C5 Instances. What's that about, what's it mean for customers? Can you explain a little bit more on that one piece? >> Sure, we're delighted to see Amazon deploying the C5 Instances. It's based on our latest technology in the Xeon product family. We call that the Intel Xeon scalable processor family. It includes, it's based on what we call Skylake technology or code name Skylake. There's a lot of innovations in that processor and that platform that are specifically driven by the needs of high performance computing. There's something called AVX 512, which is a doubling of the vector width. Means that every core can actually do 32 floating point, double-precision floating point operations per clock. That's tremendous, tremendous compute capability, in a 2X over the previous generation. On the memory bandwidth side, which is another huge factor for high performance computing applications, like 66 percent increase in memory bandwidth. So it's a balanced platform, and we're seeing improvements in high performance computing apps of anywhere from 1.7x sometimes almost up to 5x improvement in going from the C4 to C5 Instances on a per note basis. >> This is really going to enable a lot of action. IOT, tons of great stuff. >> Absolutely and as I talked about that range of HPC and you know, what fits and what doesn't fit in the Cloud, every generation of technology, what fits in the Cloud is growing, and C5 is another important step in that direction. >> Bill, thanks for coming on the Cube. Bill Magro, Chief Technologist at Intel, HPC, high performance computing. The Cloud is one big high performance machine in the sky, wherever you want to look at it, really great opportunity at enabling all new use cases, doing things for society benefits, and customers. Great stuff here, Cloud impact is significant. IOT to the Cloud. This is the Cube, doing our share here at AWS in Las Vegas. We'll be right back with more coverage after this short break. (electronic music)