AMD’s Lynn Comp on Trends in Cloud Computing

foreign speaks we're talking about cloud computing what the current challenges are and how to best handle those challenges to discuss that I'm joined by Lynn comp the corporate vice president of server product and Technology marketing at AMD Lynn very good to have you with us this morning thank you so much James I really appreciate the opportunity you know obviously AMD is an extremely high profile vendor we don't necessarily think about AMD specifically as a cloud computing company itself uh but you're obviously a cloud expert how does that work out you're working for AMD but you're a cloud expert give us a little background on that if you would store what you'll find uh most of the semiconductor companies is that you have people that are extreme product experts and at the same time you have customers that are looking to leverage what's in that product but they're creating their own products around it and so those roles applications Engineers software support software enabling you tend to engage in that solution level a little bit more and the thing that's brilliant about what AMD strategy has been is to collaborate very deeply with the cloud customers and those Cloud companies to really understand what kind of solutions do they need to take their business to the next level of opportunity because Cloud continues to morph and grow and it's been happening for 15 to 20 years now okay so it makes perfect sense so all right the the cloud computing Market there's a lot going on in Cloud these days I mean we we hear all the headlines about artificial intelligence and generative AI these days but in fact the cloud Market is still rapidly growing rapidly evolving there's there's you know hybrid cloud and you know Cloud native is is coming on strong and as you survey the cloud computing Market what are what a couple of Trends driving the sector here in mid 2023 so I think that there's really two to three that I would call out one of which is this reminder that Energy Efficiency power consumption getting the most out of your resources we've read some articles about some of the data center and the colos that are being constrained for power delivery whether in the US or in Europe and then the cloud cus the cloud consumers are wanting their services as close as possible so if you have a constrained environment and you can't route more power to that data center how do you get those new customer services supported if you are their cloud provider so really going from the practice of build out as quickly as possible to the thinking around how can I get the most out of what I have built out or how do I get the most out of my hard constraints such as power delivery physical space things like that I think that's one really big thing if you're a cloud customer as well as a provider the other one you mentioned it is AI and we are really just at the cusp this is an inflection point and I would argue we're really just at the beginning and the reason I say that is if you think about when personal Computing was just starting a lot of the use cases were researcher driven and a lot of those use cases were a little bit of hobbyist driven like how can I work with this this new Computing technology that isn't a Mainframe and yet it exploded in terms of commercialization opportunities as more and more people figured out how do I use this I think AI is really at a very similar inflection point we are just at the cusp of all of these commercialization opportunities as a result of it and I think the last one is you mentioned Edge data locality getting data in the right place at the right time for the right process that continues to be a challenge that the industry is always chasing because you'll have compute that is really high performance and then you'll end up with a situation where you can't move the data to it in time so I think that's those three yeah I mean I think on the edge piece it feels like one of the trends evolving there is that more and more of the actual processing data mining the crunching even the forecasting is going to be done at the edge so the the data wouldn't necessarily need to be moved in-house it may end up there eventually but I you see that happening where where more of the the data is being crunched at the edge itself yeah and it's honestly out of necessity because if you think about the data egress versus Ingress or downlink versus Uplink by Design Uplink is significantly uh less bandwidth than downlink because the the internet was really designed to download information and so it's almost self-preservation to do inferencing and compute at the edge so what you are sending in a very constrained Uplink scenario is the most valuable data that for example can go back and help update the training databases and retrain those algorithms but you don't want to send everything the a webcam or a traffic cam is picking up because that's a ridiculous amount of information that you would have to sort through so it's really what do you have to move that has value to the next process that's either upstream or Downstream and Edge is great for that right and and on the AP AI piece I think it's also interesting in that I totally agree with you that we're at an inflection point or we're really in many ways right in the infancy I think sort of the I rated that is that you know we know that shot GPT came out last year and it really caused an explosion of interest in generative AI but there have been a lot of large companies pouring money into AI automation robotics you know AI driven data analytics for you know several years now so it's I agree that we're this we're still at the beginning the the level of investment among large companies is not actually the beginning even though I think it's been some frustration among large companies as they've poured money in sometimes it's felt like a science experiment um I'm not sure what my question is I just think that it's yeah AI is the inflection and yet companies have been building it for years absolutely absolutely and but you know you have to think about it like the internet when did the Internet really start getting built out by DARPA and then you know you had ethernet tcpip and other things come in that really just cause the thing to break loose not to mention GUI browsers so I think a lot of the breadth chat gbt was brilliant because it gave everybody a way that they could experiment with this AI thing without having to be an expert in it and that I think was brilliant because so many people got excited about it and of course you have cautions and concerns as well let's let's look at the the cloud market and um there's a lot of companies actually still struggling with Cloud I mean I remember covering clouds say back you know 2013 2014 and it seemed like we would have had Cloud totally figured out by now in fact there would be no even no no more data centers they were saying in the early days of cloud data centers are going to go away turns out that hasn't happened data centers are very much with us and and cloud and companies are still moving to Cloud even after all these years as you look at the market I mean what are some of the common problems that the companies are having and what what advice might you give them so I think one of the main things with Cloud that a lot of people are grappling with that quite frankly is no different from 2013 it's really workload placement you know where do you run certain workloads and what is your strategy for what is developed in cloud and stays there what is developed on-prem and stays there and then what is really successfully developed on-prem and has to go to Cloud because it's a scale out function you can see a lot of companies that are being very very public about the concept of repatriation born in the cloud coming back to be on Prem and you've seen this James where it's been happening ever since 2013 2015 where you've got small companies they grow really big they realize I have optimization points that are more specific I have the ability to have hardware and locations more optimal for my application and that happens after a certain level of success for their business so I think that that's one Trend that's always been there it seems to be on a journey back to be considered on-prem a little bit further with some of the different buying mechanisms and business plans that we've seen from Hardware providers where customers have more options to do pay as you go but that one I think is is a really big inflection point that stayed there and it's never going to stop because people are always making decisions about what's best for their business sorry interrupt I can ask you a question about that the repatriation I find to be fascinating I mean I never would have thought that there would be the concept of repreciation but I do hear about it actually fairly much yeah uh you think it's mainly a cost consideration what drives that mostly I believe that it starts with cost because like think of it like an automobile if you're not going to be using your car very often and you can take public transport or a taxi because the amount of driving that you need to do is relatively low right then you don't end up feeling like you're going to need a car think about people who live in New York City or downtown San Francisco sure however if you live where I live which is Austin Texas there's very little options for public transportation and you have to have a car and so the amortization of owning a vehicle versus paying a taxi or an Uber it really is that kind of break point right and repatriation has been happening for many many years I've seen companies that were game companies that were born in the cloud and they were Cloud native workloads but at a certain point of of scale it became better for them and more optimal for them to deliver their game service with their own Hardware in the most optimized locations and colos so as part user experience and it's part cost and you only have that combination when a business has hit a certain level of growth or a certain level of maturity okay all right so I interrupted you let me let me go you please continue with the idea confidential Computing is another one that I think is a big inflection point as more types of Industries go into Cloud there's a question about security within the cloud security as you're putting things into the cloud most companies really do trust the security of the cloud which is great but there's a lot of responsibility that you still have to maintain for how you secure things going in how do you secure services to your customers so AMD for example has a capability we call Infinity guard and the way that that works is it's hosted in many of our Cloud customers instances they have customers and Healthcare as well as financials and Manufacturing and Telco and that customer holds the keys to whatever's going on in that VM there's very little performance impact to using it and so the customer can be really clear to meet regulatory requirements of privacy or needing to secure the data and the cloud customer can say I give you this capability it's not impacting their cost because the performance overhead is really quite low or negligible and at the same time both of them win the healthcare can take advantage of cloud and their scale cloud service provider can get into new industries that are highly regulated they both have an audit Trail okay and and you use the phrase confidential Computing could you give us the nutshell uh definition of that it's it's essentially knowing that there is going to be no way to get access to the data within a session and be able to infer from that private data Healthcare data financial data so it's protecting um access and readability and visibility so think about it like an attack surface as well as the contents within that session gotcha okay great yeah uh I want to talk about you you've mentioned some of Andy's role in the cloud Market did were you going to add one more thing about the challenges are you good on that I'm good on that thank you uh let's all right let's let's drill down to amd's role in the cloud Market how is AMD serving Cloud customers we don't necessarily think about AMD specifically as a cloud company how is it helping the cloud Market so first of all one of the things we just talked about with This Confidential Computing capability be in innovation of infinity guard what's really helpful about it is it doesn't require the customers of the cloud service provider to change their software to use it and so it's very seamless and again it helps our Cloud customers be able to get into more markets and service more Industries also helps their customers take advantage of that scale the availability of resources I think the next thing that has been really helpful is direct but architecture has hyper helpful in being able to provide completely software compatible architectures that meet different workload points so for example with our last generation third generation epic we had the standard performance per core Milan CPU and then we also had workload optimized version of that Milan X which added additional memory in package and what does that do for you there's a number of workloads that really benefit from high memory and being able to take advantage of high frequency so Eda or hosting Manufacturing workflows in the cloud I think the the other thing is the chiplet architecture has allowed us to do things that are more Cloud native so we can be able to scale out and hit maximum Obsession density with the same core architecture maintaining the same compatibility and many of the customers or many of the other suppliers are telling our customers you have to give up software compatibility to be able to get Cloud native and AMD is offering a no compromise you don't have to compromise you can have Cloud native and compatibility we're also doing power efficiency optimization points and in the Next Generation we'll have four different optimization points for the cloud to be able to Fan out and offer things from Cloud to Edge in whatever service profile but it's all software compatible no compromise on that hmm definitely a market for that there's no question about it um all right well I think the big question that many companies are asking is what is the future of cloud because they want to get ready for it now uh as you look ahead what kind of tech or business Evolution do you foresee in the in the future of the cloud Market I see a lot of evolution around location and sovereignty um when I was presenting at an event in Europe a couple months ago I was leaning hard on Energy Efficiency and energy profiles because amd's Energy Efficiency and consumption is world class and I had somebody take me aside on the European continent and tell me you should really talk more about security what surprised me because when you look at Energy prices you would think counterintuitively that would be where they would want to go and so I think sovereignty and continuing The Innovation path there is going to be really really critical Edge everybody is facing the situation of getting the data in the right place to not stall progress Edge is one tactic to be able to do that I do also feel like there's going to be Innovation around bandwidth we're at a point if you look at some of the an award the conversations that were happening last year we have so overblown compute relative to our ability to move the data there's going to be some balancing that I think is going to happen there as well well I mean I I definitely see the the problem with the bandwidth issue it's a real bottlenecker how do you think that will get solved so there's really only two ways to do it you have more Lanes or you move the data closer and you have bigger pools and so it's really just I think it's an optimization point of understanding your data processing needs and planning your storage tiering accordingly there are some people who do talk about um you know composable Computing and composable assets right that's I think an intriguing concept I think that location and distance could be really important in that in that whole framework and architecture but edges I believe one step towards composability because you really want to think about if your storage is not in the same site as your compute it needs to be close so right multiple campus I think will be will be part of that huh interesting yeah Lynn I think you you said it that's a lot of insight I definitely learned some things um thank you so much for sharing your your expertise today please come back and talk to this again sometime thank you I would love love too thanks so much James

This transcript was generated automatically from the video's captions and may contain errors.