00:05
Well, hello, welcome back. Welcome back for a great day too. We've got a packed day for you guys today. Uh Great agenda. Uh We've got DJ Ricky out back, rocking it. We've got uh our rotating media hasn't failed yet, which is great. Um I'm really excited to be kicking us off this morning. Uh Talking about A I, as Lynn mentioned,
00:24
this is one of the most topical uh areas of uh well, really the last couple of years, I'm sure it's affecting uh every one of you in your industries. It's affecting, you know, how you're thinking about building better customer experiences. It's affecting how you're thinking about. How do you get more efficient? How do you automate some of the things that you're doing?
00:40
It's probably affecting how you're thinking about innovating and building cool new stuff. Um You know, this is one of the most exciting times for us uh for me anyway, uh here at Pure both because of what we're doing, how we're using A I in our products. You saw some of that yesterday with cause and the innovation showcase. Uh but also the customers that we're working with.
01:00
Uh we've been working with leading customers in the A I space for over eight years, customers uh large and small from small research groups to some of the largest technology companies in the world such as meta powering their research super cluster environment and everything in between. And what I want to do this morning is spend a few minutes,
01:17
sharing some of the learning, some of what we're hearing and seeing from the most successful customers that we work with in this space. And you know what we see is first and foremost, you're under a lot of pressure, you're under a lot of pressure to move fast. You're under a lot of pressure to show value. You're under a lot of pressure to keep up with the high pace of change in this area of
01:36
technology. And yet, you know, doing a I, as I'm sure you're being asked to do isn't just one thing you get a lot of strategies, you have a lot of options to choose from uh all the way from. Hey, do I want to go uh train foundational models, uh build them from the ground up myself. Uh Maybe I want to go take the models that exist and,
01:53
and uh specialize them, add some more domain specific data and get more specialized results. Or maybe I wanna go focus my efforts more on how do I go use the models that are out there and better connected to uh the data that sits within my own organization, the data that's gonna drive uh you know, more specialized results, more specialized insights, the net of it is, you've got a lot of strategies,
02:14
you've got a lot of options you can choose from. But at the end of the day, the thing that ties them all together is they're all reliant on data. You know, a year ago you'd read in the news about all the innovations that are happening in the A I space and all you know, all the work that was happening was on the models. It was on improving the models, improving the algorithms.
02:32
If you actually go talk to people in the space today, all of the work to improve the state of the art and the technology is in improving how we work with data. And so what this brings us to is the the primary factor for success is having a good data storage strategy to to go uh tackle this. And this is where pure uh is uniquely set apart. We're the only ones in the industry that can
02:53
offer uh you a complete uh data storage strategy for all of your A I needs. No matter what part of uh that spectrum you lie on, no matter what strategy you choose to pursue. And this starts with our platform. You heard a lot about this yesterday. You heard about this from Charlie and Sean and uh Prakash and the other Sean um a single uh consistent unified platform that's highly
03:15
flexible and, and tuned and optimized for uh all of your data storage needs. For A I. We then go take that platform. We work with industry leaders such as NVIDIA. You'll hear from them in a second uh to build those into valid designs and reference architectures. We take those reference architectures and we go work with a broader set of ecosystem partners
03:34
uh to build end to end full stack solutions so that you can, you know, start focusing on deploying A I solutions and getting A I results as opposed to thinking about A I infrastructure. So I want to take a few minutes and unpack this. Uh So let's start with a platform. There we go. Um When we think about a data storage platform,
03:56
tune for A I, um there's really a couple of attributes that rise to the top number one performance, that's probably no surprise. Uh performance is important no matter which of the A I strategies you're choosing to implement. Uh Number two is enterprise readiness, right? As we see a lot of these A I projects turn into production environments as they go mainstream. Well, it only follows that A I infrastructure
04:17
has to go mainstream as well. And you know, this is a place where you're just not going to get there with a litany of HPC science projects. It's gotta be enterprise ready, it's gotta be uh production grade uh and then certainly container readiness and flexibility as well and we'll touch on that uh on those elements as well. So let's start, let's start with performance
04:35
and talk about what matters in A I um you know, I started by saying, hey, you know, a I isn't one thing, there's lots of different components to it. Uh And as you're working these projects, they're fairly complex. You might be ingesting a lot of data. You go take that data, you're gonna go work with it, index it, curate it, label it uh at some point,
04:55
you're gonna go train on it, feed it into GP US. You're gonna want to go save that work. As you're going along, checkpoint that work, you might then go take those models and deploy them into your uh operational data in fencing complex data workflow. Each one of these steps demands a different type of performance, right? You might need a high right bandwidth for some
05:14
of this high read bandwidth for other parts, really good metadata performance, mixed read, write activity, low latency over large pools of data. All that cold storage needs to go away. The point is you've got lots of different elements of performance that you've got to go hit on. And more importantly, you've got to have high levels of performance for longer than you know,
05:32
than it takes to clock a hero number and post a benchmark. You've got to have high performance in a sustained way when it actually matters. Um And this is where pure is uniquely differentiated. We're the, we're the only ones in the industry that can offer the highest levels of sustained and all around performance that uh your A I projects uh really demand.
05:51
Uh Sean C A uh drew an analogy showed a slide yesterday of a race car. You know, I think of it uh somewhat like that, right? We've got a lot of competitors uh in the, in the A I SPACE uh that are one trick ponies. They, they've built a drag racer, drag racers are pretty fast off the line. Uh But you're not going to win a race with that because it turns out that other things are
06:08
important too, like turning left, turning, right cornering breaking. Um You've got to have high levels of uh sustained all around performance and this is what we do all day long. So you need lots of different types of performance, but going mainstream means that you can't have a dozen different one off solutions to get that performance.
06:27
You're already managing way too many silos in separate islands and different products. You know, today, this is the last point in time where you don't want to be out anymore. Um As these projects go into mainstream production environments, um you know, it also means that all the things that we think about in terms of, you know, our traditional workloads, uh security, reliability availability,
06:47
these all become uh critical as well And so this is really not a place for, you know, HPC science projects or uh unproven Johnny come latelys. Um you know, in this environment, the niche players who will sacrifice all of these uh properties, reliability, availability, security in the name of performance uh simply aren't gonna cut it and then efficiency, right. Your time has never been more valuable.
07:11
Uh I'm sure power space have never been uh scarcer. Uh And so the importance of efficiency is really coming out uh in this uh age of A I as well. All right, number three container readiness, right? So as these uh environments and, and really these application stacks go into production, uh the need to be container ready uh has never
07:31
been more apparent whether you're, you know, building environments to support data preparation or deploy these models or, you know, all of the measurement that goes around them. Uh At the end of the day, these are um you know, these are applications, these are environments that are highly aligned to container stacks. They need to be, you know, iteratively updated as you're experimenting,
07:49
they need to be scaled very quickly, they're highly aligned the applications that is uh to the open source community. And so then it stands to reason that, you know, containers are a critical part of this environment. Um There's simply no A I strategy without a valid uh container strategy. And then last but not least uh flexibility. Um And I'm not going to go down the path of,
08:10
you know, telling you how many decades I've been in the industry. But uh this is one of the, you know, fastest moving, uh uh you know, uh trends in technology I've seen. Uh and this is the last point in time where you want to lock yourself into a rigid infrastructure, uh, decisions that aren't gonna grow aren't gonna scale,
08:26
aren't gonna allow you to adapt to the rapid pace of evolution. When things are changing, you know, week by week, month, by month, quarter, by quarter, the value of optionality, the value that you have to be able to flex and grow and change and evolve in different ways has never been greater. And we do this with our core platform. We do this with our business models.
08:45
You heard a lot about that yesterday from Prakash with our evergreen one for a Isls. So if we net this out and we look at the platform, uh you know, we're positioning our customers, we're positioning a lot of you uh with a complete platform that gives you all the flavors of performance that matter for A I driven by purity driven by our flash systems. Uh with the enterprise hardness and readiness
09:06
that we're known for uh the consolidation, the unification uh you know that we drive with a single operating system, uh extending that with pure fusion, certainly with container readiness. All right, the industry's most complete container storage platform with port works. Uh And then everything we're doing of evergreen uh drives the ultimate inflexibility.
09:27
All right. So uh we hit on the platform uh and that's really the bedrock, that's the foundation, that's the starting point. We then go take that platform. We work with industry leaders such as NVIDIA uh to define all of the key reference architectures. We've, we've been working with NVIDIA for uh seven or eight years now.
09:43
Uh on all of the key reference architectures from air to base pod to OVX, you heard yesterday from Sean Hansen. Uh You know, we'll be part of the super pod reference architecture later this year. And the way to think about that is now, you know, we've got the complete set of reference architectures with NVIDIA to help meet your needs, large to small,
10:02
whether it's training, whether it's inference all across the board building on these reference architectures, we're now working with a broader set of ecosystem partners to build end to end full stack solutions. And we're going to target those at key verticals, key use cases whether you're in financial services, whether you're in the Telco Space.
10:19
Um You know whether you're in healthcare and life sciences and many, many more uh to do this, we're working with key partners across the rest of the space, whether they're infrastructure partners, software, uh partners. Um you know, looking at things like scheduling and orchestration, data management, data preparation, vector databases, a whole slew of things.
10:37
Um You know, I'd encourage you to find out more about this by joining us in the breakout sessions. Uh But I'm really looking forward to rolling out more of these solutions uh as we go through uh the course of this year. Uh So to help actually talk uh help uh uh talk us through some of the partnerships and uh really the, the uh ecosystem work we've done with NVIDIA.
10:56
Uh I'd like to welcome out uh to the stage, uh VP and General manager of the DGX Systems uh from NVIDIA, uh Charlie Boyle. Uh but before we do that, let's first take a look at this video. OK. Yeah, big round of applause for Charlie.
12:53
Awesome. Well, Charlie, thanks so much for being uh here with us today. I know it's a, a vacation uh break for you. Yeah. No, great to be here and great to see everyone. You know, it's a first for me today, first time on a turntable. Uh So thank you for that, Rob. I didn't fall off.
13:09
Great. Well, uh look before we get started. Uh you know, I still remember the first NVIDIA product. I bought it as an old G force uh video card in the gaming system back in the late nineties. You've been in the NVIDIA for a while. Uh Now the world's largest uh company, tell us a little bit about uh your journey and uh your role.
13:25
Yeah, I mean, I, I've been in enterprise computing data centers, you know, my whole career and uh back in 2016, you know, about 8.5 years ago, had the opportunity to join NVIDIA because they were building something new, the, the world's best integrated A I system. And, you know, back in 2016, people didn't really know what that was.
13:46
But, you know, our Ceo Jensen had a vision that, you know, we needed to bring the best possible A I system to the market ahead of what people needed because he knew and the company knew that A I was gonna be huge and, you know, we had to show the world how to do it. And we've been on that journey for the past, you know, 8.5 years now with DJ X with the beautiful gold servers,
14:08
the things that you saw in that video to really help customers, you know, build that A I infrastructure, build the, you know, those A I solutions. Uh and pure was one of our first uh partners, you know, helping customers, you know, make that, you know, solution a reality in their data centers.
14:25
Yeah, and it's been a great partnership. Uh You know, you speak about uh kind of the journey uh you know, helping customers uh get to where they want to be with A I, you're out there talking to customers uh day in, day out. Uh What are the main critical challenge? The challenges that you see customers grappling with.
14:41
Yeah, I mean, you know, we're, we're here at, you know, piers conference here, you know, a lot of it is around data because, you know, data is the fuel for A I, you know, otherwise it's just a bunch of, you know, computer science and algorithms. But what makes A I useful to you is the data in your company, the data in your ecosystem that helps you build your A I applications to serve
15:03
your customers better to make all those things work. But the world has changed really rapidly. You know, when we first started, you know, working with pure and data, it was all about A I training and I had a large static data set that I had to train my model on. But it's changed so dramatically now, you know, back in those days,
15:22
it was all labeled data. You know, the the very first A I the cat detector, you know, thousands of people in a room like this had to look at the initial training set and say this is a cat, this is a dog, this isn't either. But now A I doesn't rely on that label data. It's training off of unstructured data data that's in your company. And it's learning from that information.
15:44
One of our biggest customer challenges around data has really been the change in A I algorithms. You know, like I said a couple of years ago, it was static training data and then it was unsupervised learning, you know, but now it's taking a, you know, a large language model, something you know, that's based in English. But adapting that to,
16:05
you know, the enterprises, you know, lingo their own words that they use the same word in English may mean something completely different in two different industries. And so that rate of change has really caused customers to really need to be very flexible in the way they address their data. And also look at how they do it efficiently because, you know,
16:24
I think Rob, you know, mentioned earlier that, you know, data center spaces as an all time low electrical power availability is an all time low. So doing that efficiently and being able to move as the industry moves because this isn't a jump every 3 to 5 years, it's every 3 to 5 months now.
16:40
Yeah, and we're, we're seeing a lot of the same things. Uh You know, I talked about flexibility. I talked about efficiency, uh a lot of change going on. You know, I think one of the things uh I hear a lot when I go speak with customers is, you know, we see the great promise. We see the great technology, the the components.
16:56
Where do we get started? All right. Do you hear that uh from customers? And I guess do you have words of advice for folks in terms of uh where they, where they should, uh, you know, jump in and get started. Yeah, I mean, the, you know, the biggest bit of advice, you know,
17:07
and I started telling people, you know, about this back in 2016 is you just have to get started with an A I project. And back then you used to need phd S on staff, research scientists, all these things. But now with the explosion in available A I applications, you know, large language models that you can download from the NVIDIA site or
17:30
lots of our partner sites that then you can just customize to your own business. You know, getting started is all about taking that first step doing something that may be, you know, easy for your company, an internal chat bot, you know, uh you know, every company that I've ever worked for, you know, your internal search, your internal systems, you're like,
17:52
I know this bit of information exists somewhere. I heard someone talk about it, but it's almost impossible to find. But if I have a chat bot that I put in front of a lot of my enterprise data, I can get access that I can unlock a lot of value inside my company. And it's also very safe because it's inside my corporate walls. I, you know, I don't have to worry about
18:11
letting the whole world see my dirty laundry or something that somebody wrote once, you know, it's, it's all inside. So get started with something small but they can really help your company and don't take on a project that's gonna take three years, take on something you can do in a couple of months, show a quick demo and when you get that excitement in your company, it just explodes.
18:29
I think that makes a lot of sense. I mean, if, if I were to net that out, don't wait on the sidelines. Jump in, get some quick wins and grow from there. Um You know, if uh if we rewind one step, we talked about flexibility, I think we both see that need, you know, from the data storage perspective,
18:44
we've been helping customers uh achieve that flexibility with evergreen and non disruptive upgrades. Uh really since day one. what are the strategies that you see customers uh adopting for, you know, achieving that flexibility in other parts of the stack? Yeah. And you know, I mean, I think it, it started,
19:01
you know, you said pure was our first storage partner and it started with a customer need, you know, back all those years ago, people started to cluster their systems and they said I need multiple users to be able to use the same system. Well, now that's just been multiplied 100 times 1000 times. Now that, you know, people want that flexibility.
19:21
They want to know that the thing that they're buying for their infrastructure is not only going to work today, but it's going to serve them for the long term. So, you know, look for something that has a long term reference architecture. You know, we started with pure with the A E reference architecture. All of that work turned into the program that we now have today as base pod and you know,
19:41
as we talked about in the in the future Super pod, but it it's building that durable infrastructure because all the customers, you know, that I talked to even, you know, when we did GTC earlier this year had uh a great customer on stage with me and his advice to everyone was just like when asked the question, how do you know how much to buy? Now, the question is, I know I'm always gonna need more,
20:02
but I've got to get started, but I have to get started with something that's modular that I can grow with. Because what's really changing the industry now is people aren't buying a training cluster and an inference cluster. They're looking at one thing that serves their whole business, you know, and we've coined the term for that, you know,
20:20
as an A I factory, you know, and I always think back to Saturday cartoons, you see the big factory and there's this huge lever in front of it, they turn the factory on and off. But you know, as I look what an A I factory is now, well, it still is that same big lever in front of it and you pull the lever one way, it trains your new models fine tunes those you push it the other
20:39
way and it's doing inference on all those models. So, you know, invest in an architecture that can grow with you that you can add to and that's going to give you the flexibility to deploy that A I factory and be successful and be able to grow in the future. That that makes a lot of sense. And, you know,
20:55
I think as we think about deploying um you know, that those factories and kind of the uh scalable infrastructure you talked about uh and I talked about as well, uh different strategies that people can take uh as they get started. Uh Whether it's, you know, do the foundational training, doing the fine tuning, uh focusing on inference and Rag. Um you know,
21:13
what uh you know, what advice would you give or how would you guide uh folks that are kind of waiting to get in and, and uh looking to get started uh as to how to think about the different strategies. Yeah. And you know, you, you, you mentioned R A and that's probably the biggest advancement for all of enterprises in A I because prior to Rag and R A as you know, retrieval, uh augmented gen A I,
21:38
you had to train a model and you had to think that that model had all the information it was ever gonna need. But that fundamentally breaks down because I can train a very large model and it could fail on a very simple thing. You could ask it a historical question, knew that. But if you asked it what the weather is going to be tomorrow in Las Vegas,
21:59
it wouldn't know it was trained last year. And the ability for A I models transformer models to be able to retrieve new information is really what unlocks, you know, value in the enterprise because you have a fundamental feeling about your company about the way data is structured and how information should flow. But you're creating new information every day, you can't retrain your model every second.
22:26
When there's new information, you need a foundational model that's applicable to your business. But to have enough connections to your enterprise data store, to be able to pull new information, you know, something that just happened in Pur's keynote today as a company, you know, as soon as that slide deck is posted information, there should be accessible to
22:47
everyone in your enterprise. You know, rag really unlocks that capability. So when people are looking at getting started, there's so many great rag examples out there. I know pure has done some uh as well for customers because you know that that's something that shows super easily inside of a company, but really unlocks value quickly for you so that you can expand and get better and get more detailed.
23:09
It's kind of like the difference between training a doctor and then giving them access to the patient's charts, they can make better decisions. Yeah. So uh let's let's maybe bring this back to pure NVIDIA. We've worked together for 67 years and collaborated on a number of reference architectures. Uh The latest one being Super Pod coming later
23:26
this year. Can you share a little bit with the audience? Uh why the Super Pod announcement is so meaningful and, and exciting and, and really the impact that you've seen Super Pod have on, on customers in the enterprise. Yeah. And you know, as we've evolved over the years in A I infrastructure to look at what customers needed,
23:44
you know, we started with very basic things. How do you connect things together? You know, and then we started needing to be more opinionated. How do I make sure I get enough training, performance and what customers really got to was they needed not only a system that could give them those things, but they actually wanted guarantees around it and they wanted to know,
24:05
you know, the the biggest question I would always get asked is, well, NVIDIA, how do you do it because we spend, you know, hundreds of millions of dollars, five billions of dollars with all our chip development on our own internal clusters. And we do that to make sure that they work well for customers when they get them. But the most common thing,
24:23
you know, theme, you know, years back with customers is. Well, how do I get what NVIDIA does? I see these great benchmarks. I see these world records. I see all this mo perf, how do you do it? And we came up with this idea of a super pod and a certified solution, which is, you know, NVIDIA hardware,
24:39
NVIDIA networking, great partner storage, like pure and it's not only tested that it works, but it's in, it's a turnkey solution. There's a full recipe. Our professional services comes out, you know, pure professional service would come out. And when you're left with the solution, you actually get a performance guarantee on it.
24:58
When you do a software update, you know that the performance is always going to get better. The same thing that you're running internally is the same thing that we run internally and that hundreds of customers use. And it really allows you to just get your job done. And that's really the goal of Super Pod is any A I,
25:15
you know, new solution that may come up on to market. It's going to run on your Super Pod, new inference. And I talked about A I factory that same Super pod is doing inference now and it's doing training. And so as a customer, you want to have something that's always updated, always secure backed by NVIDIA backed by great partners like
25:32
pure and so just get your work done, you know, you can deploy that super pod as enterprise it and never have to really get deep into how to run an A I system. You need to know how to run a high performance computer. All of you do that today, but we take all that guess, work out of it by building that by putting a ton of engineering work into it and we keep it up to date.
25:54
You know, we do quarterly releases, six month releases combined with our partners. So, you know, if you get an update from us, it's gonna keep working. It's a ton of work on our side. It's a ton of work on our partner side like pure, but it leaves our customers in a much better place. Yeah, but at the end of the day, you get something that really works awesome.
26:11
Well, thanks for joining us today. If I were to net this out, get started quickly, notch some quick wins. Focus on flexibility, reference architectures uh with uh key partners uh such as uh yourselves are, are gonna help you the recipe for success. Yeah, excellent. Well, thanks for joining us today, Charlie. Um And before we move on,
26:30
uh I'd like to uh introduce another video uh of a breakthrough award winner. Uh Can we roll the video? Softbank Corp is one of Japan's most revered tech conglomerates, whether it's exploring new use cases for artificial intelligence, developing its own large language model for the Japanese market or driving groundbreaking
26:56
telecommunications research and development. Softbank needs uninterrupted data management and integrity with a pure storage platform. Soft bank gains a flexible and reliable storage environment that ensures 99.999% uptime and non disruptive upgrades while reducing its total cost of ownership and making its business more sustainable.
27:19
Congratulations to Softbank for winning the pure storage greatest of all time breakthrough award for a PJ.
