 Hi there and welcome to another edition of Tuesdays with Corey. We've been on a little bit of a hiatus mostly thanks to Rick not doing his job. So we'll put that maybe aside for now. But today we have a fantastic show. You see no laptop today. Today is going to be just a talking show, a talky as we say. And we will be talking about HPC. And so why don't you tell the audience what you do, what you're working on and we'll get into some questions. And who you are of course as well. I should have introduced you but I'm going to let you do everything now. My name is Evan Bernes and I am the lead program manager at Microsoft in our specialized computing division for our high performance computing offerings. And I'll be talking to you today about our new offerings for HPC called the HB and HC series virtual machines. Awesome. So right off the bat what do these letters mean because I'm going to act like I don't know. Yeah. What are these letters? What kind of idiot came up with these names? That idiot would be me. Oh no. I thought it was me. I'm sorry. You're going to say me. So the H was me. Let's be clear. But then the additive, the additive C and B was you. The B and C was me. You've made it better. Let's say that. You've added to make it better. The original naming scene was broke. Sure. I'll take that. There you go. There you go. There you go. How about that? All right. Go explain what each one of these cues are and they've recently long. I mean they recently just came out. We've just gone into GA. We're just exiting a super successful preview period and we're in GA now. Okay. Okay. So the two ones, HCHB. Hit me. What are the... So C compute B is big. C is for compute. B is for bandwidth. Bandwidth. Yes. I knew it. If you look at HBC workloads, it's really important to get super specialized in terms of what are the characteristics of the underlying technology that actually drives the applications forward from the standpoint of performance, scalability, and cost efficiency. Got it. If you look at the broad brush, you've got a pretty even division between workflows that are driven principally by memory bandwidth, how we can get data in and out of DRAM to a local CPU, and workflows that are driven by dense computation, a lot of flops, a lot of vector math. Right. Right. And that's really then just the speed of the CPU itself that's going to matter. That's correct. Got it. So maybe it's surrounding caches and so on. So memory-based workloads, this is where you've got large CAD files and you're sort of working against them. Like what are some of the examples? Was that right or is that just like totally... It certainly can be. Oh, that's nice. That's a nicer response to me. Certainly high-end scientific and engineering workloads definitely are going to take advantage of this. Of that memory bandwidth. Yeah, the workload you'll see that loves memory bandwidth all day long is fluid dynamics. Got it. So any manufacturing organization, oil and gas organization, public sector research that's looking at things like black hole, supernova, plasma out there in the galaxies. Huge amounts of memory bandwidth will take as much as we can give them. And we've prioritized that first and foremost so we released... And we're like leading in this, right? By a long shot. By not even close. Yes, by... If you look at some of what else is happening... This B is much bigger than other B's that we've seen out there. That's correct. If you look at the competing clouds out there, we typically can do about 1.5 to 2X the memory bandwidth that they can do often at lower cost. And this is really part of what differentiates Microsoft with our specialized computing division, which is we don't just do general purpose offerings, we do super specialized stuff for the customers who really need that level of specialization. That's awesome. That's awesome. And so that, you said that's a good oil and gas example, some manufacturing as well, obviously anything where there's fluids involved. Yep, absolutely. If you think about, say for instance, an airline manufacturer, jet fuel, wind going over the wing of a... Yeah, got it. Cool. Got it. Now, super floppy workloads. Super floppy. I've never heard that. I'm going to steal that. Well, now you have. It's a type of workload. There you go. And it's focused on flops. Super flops. There you go. Yeah. Super flops. They don't let me talk to people often. There you go. There you go. So tell me what the types of workloads that we're seeing during the preview, right? You saw a bunch of these during the preview. And so what are some of the great examples of that type of workload? Well, one workload that really, really likes it is structural mechanics. So structural mechanics is sort of the complimentary angle to what you have with fluid dynamics. So the world is principally constructed of fluids and solids. You've got gases in there, but gases from a simulation standpoint are just another form of fluids. So if you think about, for instance, car crashes, where all the car companies, they have to prove to safety regulators that if my car hits a wall at 60 miles an hour, what's going to happen to the structure of the car? What's going to happen in terms of the force that's exited on the passengers inside of the car? Very cool. If you prove to regulatory agencies, you're making a safe product. So these workloads are super duper important when you think about airline engines that you need to show that if they blow up, they're not going to cause the plane to fall out of the sky. Cars crashing into walls, things of that nature. And then when you can combine them, where you get really groovy stuff, is when you can combine CFD workloads with structural mechanics workloads into a single simulation or an end-to-end workflow, where it's doing what's called multi-physics, where it's simulating all these different classics. All together. All the physics, all together. So like the molecules of a car in a crash coupled with the wind going over it and sort of how that plays out, something like this. You got it. Nice. And now with the interconnect between these, how important is the interconnect between these nodes? So like I'm assuming that if you're doing some sort of calculation like this, either the fluid dynamics or the sort of structural things, you're doing more than one machine is involved typically. By definition, the HB and HCC series are all about multi-node runs. Not just a couple of nodes, a lot of nodes. So Azure has the cloud's first deployment of 100 gigabit networking. And we go a step further and we have 100 gigabit InfiniBand networking with SRIOV. And why that really, really matters is that means… Don't send that network right into the CPU. Not just that, but it enables customers to use the InfiniBand. We have an Azure exactly as they do in a bare metal deployment. So if they spun this up themselves, it would look just like this. It looks exactly like it. And the networking we have is literally as powerful as the interconnect you find in the world's two fastest supercomputers. We brought it into Azure and now we're deploying it boldly. My brain being one of them and what's the other way? There you go. Yeah. My brain. Okay. And that brain is what stopped me from thinking of what my next comment was going to be. So that's awesome. GA. So we just went GA. We just went GA. Everyone can go do this type of crazy fluid dynamics, structural processing, Yep. Super computing work. Anyone who wants it. Anyone who wants it. Just fingertips, spin it up per second billing, same jazz, just the world is your oyster. You got it. That's awesome. Yep. Yep. And so I'm looking for… I'm going to go play with this right now. I think I'm going to go do some… I've got to find some sort of fluid that I'm doing some dynamics against, but I'll figure it out. So that's awesome. People can go use it. Any other things that you want to call out about this? Anything else that I should have asked that I didn't? I think the big thing for people to know is that Azure is super committed to this space. We are not just doing the HB and the HC series now, which is what you can find here in 2019 in GA. This is part of a pattern you're going to start to see from us in terms of doing new technology deployments year over year and sometimes even faster as the technologies come to market. We… This is a… A lot of these technologies have an insatiable appetite for performance, for speed, for scalability. They want to be able to run stuff at scale that's orders of magnitude beyond even what's available today. So there's no such thing as fast enough. Right. To be able to serve those customers. Just going to keep… We have to be on the cutting edge, and that's what we are committed to doing in Azure. So we can't talk about what we're doing later this year just yet, but we're super jazzed to talk about it in a few months and show people the advancements that are coming even beyond HB and HC today. That's awesome. Do you have any sense of how many additional letters you're going to be using for these sizes? Like six, seven, eight. You don't know yet. You can't reveal that. I like maybe just going in alphabetical or like B, C, D, something like that. Nice and simple. People can remember that. Super boring. It's not… What letter would you like? Like a Klingon letter? You know, for the longest time I wanted a Z. Yeah. Like for something like super special… Zoro. The Zoro skill. Yeah. That's nice. Yeah. Okay. Yeah. Right here. All right. I think we're done. Okay. I mean, with that, once you start talking about Zoro, I think maybe we're done. So thank you, Evan. This is fantastic. You're welcome. I'm really excited. Thank you. And you guys should all go check this out. In the meantime, if you've got any questions on anything that we talked about here today, any questions about HPC, about what we're doing in this business, in this space, hit me up on hashtag Azure TWC and we will make sure the right questions get out. And I don't know if you have a… If you want to share a hash, you want to hashtag this out. Are you a big Twitter fan? Yeah. You can find me at Evan Burness or at Evan Burness on Twitter. Okay. Perfect. That's an easy one. So hit us up. Ask some questions. And thank you so much for your time. Have a wonderful Tuesday. Thanks, everybody. And we'll see you again kind of soon. Maybe another big break because Rick again is on a joy ride around the world. Once he's back, we'll be back on target. So good to see you and hope to see you again soon. Bye-bye. I have no real assurances that this isn't like for an Azure outtakes real. It could be. I mean this is all… This right now will be… If you do something funny, it will be on an outtake. Great. But right now what you're doing is notably not funny. Okay. Or interesting. Good. Just FYI. Yeah, it's good.