In The Arena by TechArena - Couture Silicon for the Cloud with Ampere's Jeff Wittich

Episode Date: November 21, 2022

TechArena host Allyson Klein talks with Ampere Chief Product Officer Jeff Wittich on rise of Ampere fueled computing in the cloud and why Ampere's lineup places it in an excellent position for the next wave of cloud growth.

Welcome to the Tech Arena, featuring authentic discussions between tech's leading innovators and our host, Alison Klein. Now, let's step into the arena. Welcome to Tech Arena. My name is Alison Klein. Today I'm joined by Jeff Wittich, Chief Product Officer at Ampere. I'm really excited for the conversation today. Welcome to the program, Jeff. Thanks, Alison. Great to be here. So Jeff, you're a longtime player in the tech industry and have a tremendous amount of experience with selling different architectures to large consumers of data center infrastructure. Most recently, you're the chief product officer at Ampere. Why don't you give us a background on Ampere and why you decided to set your sights

on this new architecture? Yeah, absolutely. So Ampere has been around for about five years now, and we really set out to create a processor that was specifically designed to meet the needs of the cloud. A lot of us have been working with cloud customers, with cloud technologies for even the last 10 or 15 years. And it's obvious that a lot of things have changed. The architecture has changed. The software world has completely changed during that time. But what hasn't really changed

is the underlying hardware itself, especially the underlying microprocessor that runs the cloud. And so at Ampere, we really set out to correct that. We set out to build something that's new, something that's able to deliver performance in the way that the cloud requires it, that's scalable the way that the cloud requires it, that's scalable the way that

the cloud requires it, and especially something that's efficient in the ways that the cloud requires it. We've got some great products out there right now that do a lot of these things, and we're really excited about our future roadmap as well. You know, I think that the news has been filled lately with stories of some of the largest players in cloud adopting the Ampere architecture. And I wanted to ask you a bit about some of the things that you just talked about. What are the determinants that have changed in terms of cloud providers and what they're looking for from an architecture and what's changed in software? Because we've seen other architectures come in and try to do what you are being successful doing. And so there's got to be something different afoot.

Can you tell us a little bit about the Ampere architecture itself? And what do you think are the key characteristics of that architecture? Yeah. So I think one of the most important things strategically with what we're doing is we really took a clean sheet of paper approach and focused on the actual problem statement, which was how can we build a processor that's uniquely designed for cloud infrastructures? And when we think cloud, we're thinking pretty broadly. We're not thinking of some subset of workloads or a subset of locations. We're thinking about any place, any workload that runs in a cloud

infrastructure, which is over time becoming almost every single workload. Almost everything is moving towards this cloud type of environment. And so if you look specifically at what we did, if we take performance, for example, obviously everybody wants high performance processors that deliver the most output across a variety of workloads. But there's some nuances to that. They want performance that's predictable, performance that's consistent. What's changed with the cloud is that you now have dozens, maybe even hundreds of users and applications running in the exact same physical piece of hardware. And that means that you really do need to re-architect things at the processor level to ensure that you don't have a bunch of unnecessary or undesirable

interference between users, where users are kind of stealing their unfair share of resources, which causes a lot of unpredictability, or even that users are starting to infer the contents of other people's private data. So that predictable performance has been one key. The other is that the way that we're delivering that performance also relies on having a very large number of cores. Each of those cores are really high performance. But having 128 of them, for instance, means that our customers have the scalability they need to provide the capacity to their users. And we make it easy for them to utilize all of those cores, run them at nearly max utilization, and not see some sort of performance cliff where they feel compelled to reduce utilization to achieve max performance.

We're able to deliver max performance at max utilization all the time. And then I think the third one that's becoming really important, especially over the last 12 months, is just delivering that in a really power efficient envelope. I will say that a couple of years ago, that was perhaps the least important of the attributes of our processors for our customers. It was a nice to have, but it wasn't absolutely required. And clearly with some of the changes that have occurred over the last nine to 12 months, scarcity of power in certain places, pushback on further data center sprawl, concern around increasing power consumption, increasing

water usage in specific regions, and just how hard it is to build that extra data center or build the extra capacity on the grid to satisfy more data centers has made power efficiency a big part of a lot of our customers' calculations, where if they can deliver two or three times more performance in the same power and data center footprint, that's a big advantage. And not having to build that extra second or third data center matters. I think we can all feel that viscerally with what's happening with energy prices. And when you extrapolate that to large-scale data center sizes, you really get the impact of being able to deliver more power efficiency. Why is the

Ampere architecture such a star when it comes to performance efficiency in the data center? Yeah, I think it's a couple things. And a lot of it, it always comes back to that focus on the cloud use case and making sure that we remain very focused about what we're delivering. And what that means, quite simply, when it comes to our processors, is we have built our processors, the cores that are in our products, starting at Ampere One and beyond. We built them with this extreme focus on ensuring that everything we do that delivers more performance needs to do so without consuming incrementally more power

than the performance gains that we get. And that might sound really simple. It's not. If you focus on too many different use cases, it can be really tempting to continue to add accelerators that do this or do that, features that do something for one very specific use case, but don't actually benefit all those other use cases. And we can just eliminate all those things and focus entirely on things that deliver extraordinarily efficient performance for cloud use cases. Another thing is that, you know, it's not only that a lot of other processors weren't designed with cloud use cases in mind, a lot of times those cores weren't even designed with data center use cases in mind.

And so a lot of those cores are actually used in client products, which have power efficiency requirements as well. But those power efficiency characteristics are very different. Those requirements are different than the efficiency requirements in a data center. In a laptop, you are expected to have big idle periods. You want to just sip power during those extended idle periods. You want to be able to turn things on and off on demand, but you're okay with burning a lot of power in a short amount of time if it means that your latency decreases and you can get a job done faster and then go back to sleep again. And that's just not the case in a data center. In a data center, you're expecting to have high utilization all the time, and that's just not the case in a data center. A data center, you're expecting to have high utilization all the time. And that's the operating region to be really efficient because actually efficiency improves as utilization. working with some cloud service providers as some of your initial customers to seek their commitment

to deploy. Tell me a little bit about what that collaboration is like and how much the customer is influencing the design of where you take your architecture. Yeah. Yeah. We've been working really closely with a number of the biggest hyperscalers in the world now for a number of years. And that's been really important. I think we had a pretty solid idea of what was needed when we came in. And a lot of that came from the fact that we have a team that has spent, in many cases, over a decade working in the cloud industry and knows what some of the fundamental challenges are. But of course, we've learned more about the nuances, the implementation details, some of the specific operational details, you know, by working really closely with these big cloud providers.

And that's really why we focused out of the gate at going after the big hyperscalers and trying to win at the big hyperscalers. They're the hardest customers to win, but they also give you the most feedback and are the most demanding customers. So if we can meet their requirements, then we surely can meet the requirements of other cloud operators, cloud providers around the world. So having those years of input from the hyperscalers has been really valuable to us, not just in the current products, but also in our future ones. And that's also one of the key differences with how we're developing our products is we have a really fast cadence and we're using a lot of agile techniques, ones that really

come more out of the software world than out of the hardware world. So we can iterate really fast. We do a lot of work in pre-Silicon before we ever get hardware back because actually getting hardware is the slowest part of the process. And because of that, we have tons of iterations on our products before we even have them physically back. And because of that, we can incorporate a lot of customer feedback very rapidly. We don't have to wait three, four, five years for some future product to include a new customer request. We can actually do that in 12 months, 18 months. That's impressive. When you think about all cloud architectures and everywhere that cloud is deployed, that puts you squarely into the space of targeting enterprise as well. How is the

conversations with enterprise gone? And are they as ambitious about trying new architectures as some of the service providers? Yeah. Like you mentioned, yeah. Enterprise is definitely a key part of our strategy. As we started out, it was about getting the big hyperscale providers, the big infrastructure as a service players. And then we started to get a lot of wins with some of the other service providers, the digital service providers, social media companies, things like that. And especially over the last six months, since HPE has come out with the ProLiant RL300 servers that are based on Ampere, we've had increasing interest from the enterprise private cloud, which is a key part of our overall full cloud strategy. It's all about

public cloud and private cloud and all of the hybrid models connecting those up together. So I would say that we've seen increasing interest over the last six months, especially in things like the financial sector, healthcare, a lot of the industries that are incredibly dependent on compute for their fundamental capabilities. And I would say that the push for sustainability and the need for power efficiency has been a key reason that that conversation has sped up quite a lot, is the fact that they clearly don't see building out additional data centers or finding additional data center space and capacity somewhere else

as an option, and their needs aren't slowing down. So I would say that we're going to see a lot of enterprises adopting Ampere processors as they continue to build out their private clouds. And I'm really, really excited about where we're going with that. That's fantastic. I can't wait to hear more. You mentioned future products. So obviously, I've got to ask the question of what we can expect from you here headed into 2023. We've got some headwinds with macroeconomic forces. It sounds like you almost have a perfect product, though, for challenging times.

And so I'd love to hear what you've got in store for us in the next year. Yeah. Yeah. As you said, there are headwinds definitely in the macro view of the world. But these types of periods are typically where technology transitions occur. People have every reason to go and look for a new way of deploying compute, a more efficient way of deploying compute. It actually helps to solve perhaps some of the business problems that they have. Maybe they have shrinking budgets. They can do more with their

existing footprint with our processors, or maybe they need to reduce their power consumption, reduce the power bill. Our processors will do a great job with that. So I'm really excited about this because this could be the catalyst for that change that needs to occur. And I think as we go into the next year, we already kind of hit on the fact that I think we'll see a lot more enterprise adoption. Most of what we've seen to date from us has been around public clouds, but I think you'll start to see more and more private and hybrid cloud use cases come out. The second one is, I think you're going to see a lot more at the edge. So we have a couple of

really interesting edge wins. Cruise, for instance, is one who is using us for their robo taxis for them you know that's just a really high performance edge server that runs cloud workloads and so if we can succeed in that type of high performance use case you know there's a multitude of other edge use cases that need high performance but have constrained environments that will do very very well. So I think you're going to see some really, really interesting edge use cases. And then the third thing is we've had great success with our Ampere Ultra and Ultra Max processors. And shortly you will see more from us on our Ampere One processors, which use our own Ampere custom core. They'll go out to some much larger core counts than what we have right now. And they'll continue to deliver more and more performance, efficiency, and some cool cloud features as well.

So I'm really excited about getting those out in the market and being able to talk more about Ampere One. Fantastic. Well, thank you so much for being on the program today, Jeff. I've just got one more question for you. Where can folks find out more about all of those cool products that you just named off? I'm really excited to see more about the Ampere Ones. That sounds really interesting. But the entire lineup sounds great. Where can folks find. If you go to AmpereComputing.com, you can find a lot of information on everything from the products, technical details within that around the products, solutions that people are building out around our products, software, performance data is all on there. So go to the website. And if you don't see everything you're looking for, there's tons of contact information on there and you can reach out to us and we'll be happy to answer any questions that you

have. Thanks so much for being on the show today, Jeff. It was a great time. Thanks, Allison. Thanks for joining the Tech Arena. Subscribe and engage at our website, thetecharena.net. All content is copyright by the Tech Arena. Subscribe and engage at our website, thetecharena.net. All content is copyrighted by the Tech Arena.