Skip to Content
16:30 Video

Accelerate AI with Pure Storage and NVIDIA: Fireside Chat with Charlie Boyle

In this Fireside Chat, Pure Storage CTO Rob Lee welcomes Charlie Boyle, VP and General Manager of DGX Systems at NVIDIA, for an enlightening discussion on the current state and future trajectory of Artificial Intelligence (AI) in the business world.
Click to View Transcript
00:09
Hello, everyone and welcome to our Fireside chat series, uh with Pure uh today. Uh I'd like to welcome, uh, Charlie Boyle, uh VP and General Manager of DGX Systems from NVIDIA. Uh, you know, to sit down and, and chat with us a bit and, and trade thoughts on, uh where, uh, you know, where we sit with A I and just have some
00:28
discussions uh around, uh, everybody's favorite topic. Uh De Jour Charlie. Welcome. Thanks for being with us. Yeah, th thanks Robin. Uh really, really happy to be here. It's, uh super exciting. We're coming up on the eight year anniversary of when we first launched DGX. And, you know,
00:44
it's incredible how the A I industry has expanded from, you know, where we first started in research and, you know, you needing a phd to, uh, you know, deliver A I to, you know, our early partnerships with Pure and helping bring, you know, A I into the enterprise and today, you know, A I is just really everywhere it's on,
01:03
you know, everyone's hot topic list. How do I use it? How do I get started with it? Um, you know, and just seeing that, you know, explosion in A, I, you know, in every sector and every customer. Absolutely. I agree.
01:15
And, you know, when we first met years ago in those earlier days, uh, not sure that either of us could have predicted, uh, you know, kind of where, where we're at today. Uh II, I guess that kind of, um, you know, brings to mind, uh, you know, the first question, uh, for me is, uh, you know,
01:31
with what we've seen, uh, with A I over the last couple of years and, and the buzz that is out there today, uh, and the developments over the last 678 years. Um, where do you, where do you think we are kind of in the development cycle of A I technology? Right? Are we, you know, are we, are we in the seventh inning stretch,
01:48
are we, uh, you know, uh, down in the ninth inning or are we just getting started, uh, you know, in the 1st, 2nd, 3rd innings? Yeah, I mean, it, it's, it's been great to see the growth in, in A I and, you know, doing this day in and day out, you know, you, you think everyone's doing it but, you know, for most enterprise customers,
02:05
you know, their wake up call was chat GP T and, you know, they started seeing, you know, the power of A I and how it could really transform their business from, you know, just having basic chat bots to actually having, you know, real conversations and getting real uh answers and access to data for customers. So, you know, from an enterprise perspective, you know,
02:28
it's early but it's early and real meaning, you know, there's real applications today. There's great examples of how people can get started and they don't need to completely start from scratch. I think one of the biggest developments over the past year has been pre trained A I models that get enterprises very close to what they need. And the, the difference of, you know, not
02:51
having a pre trained model versus having one, you don't have to spend months or years training that model. The biggest thing you need to do to the enterprise is add you your own unique enterprise data to it. You know, and that's really how you customize the thing that your users see and make the model something that, you know, has the personality of your company.
03:10
Uh because at the end, you know, A I is all about data, you know, and how customers take the data they have and transform that into a model that then serves their customers, you know, but while I'm, I'm interested because, you know, pure customers, you know, have a lot of enterprise data in that, you know, what are you hearing from them and what should they start looking at,
03:30
you know, for their A I initiatives? Yeah, I, I would agree. I think uh what we've seen is very much the same 56 years, or really, even three years ago. I think, uh A I projects were seen as from the broader enterprises, almost a, a bit of a, a too big of a hill to climb. Right.
03:49
If you have to start from scratch and do foundational pre training and without a clear end in sight, it's a, it's a lot to bite off. I think the biggest change we've, and we, we hear from our customers is with uh you know, the admins of chat GP T and the focus on LL MS and the availability of uh really high quality pre trade models. It's really democratized access to the broader
04:10
technology for the enterprise enterprises can now look at how do they connect their data sources into these models into this technology, whether it's through fine tuning with their main specific data, whether it's through techniques like uh retrieval augmented generation or R A. Uh and, and being able to bring together uh you know, their operational data sets, maybe their historical data sets uh to,
04:35
to uh you know, produce better and more focused to their business results. I think that's where a lot of uh the focus is in the enterprise today. And, and I think that um I think it's real but I think it is earlier uh in that, in that journey. Um You know, I think from uh you know, from the point of view, if you're a pure customer, really,
04:55
if you're a customer in the enterprise that uh wants to get there and, and you're thinking about what does it mean to uh to deploy these A I environments? Um You know, the, what we see the our most successful customers that head down this road uh really are looking for a couple of things at a high level, right? One is um the more that you can take your
05:16
enterprise data and connect it into these systems, whether it's through fine tuning, whether it's through uh R A, uh the better you are. So number one is uh figure out what data you have and where it sits uh and break down silos, right? Uh The typical uh his, you know, the historical enterprise data environment has been very fragmented and siloed.
05:37
Uh And that's an absolute anti pattern for getting the most value of connecting those pits of data together. Um The second is really preparing uh infrastructure and, and uh we see our most successful customers uh planning and preparing their infrastructure, not just necessarily for uh you know, speeds and feeds uh but also for just as much so reliability for,
06:03
you know, these are mission critical environments, all of that stuff matters. Um And then lastly agility, right? I still think this is uh you know, a fast moving space. I think, you know, our partnership continues to, to, to move along at breakneck pace. Um Their technology is going to keep evolving. And so uh being able to be agile about your data storage uh plans,
06:25
uh your infrastructure, what those workflows look like, I think is really important if you design yourself into a box where uh you know, next year's model sets or next year's techniques don't cleanly fit into that box. Um That's gonna be, be a real problem. Um And so that's kind of what we're seeing. But uh Charlie I'm curious,
06:42
you, you work with uh you know, many other customers and partners. Um What do you see as uh you know, what do you see as kind of the top considerations from a data storage perspective? Yeah, I mean, I, I think you touched on a few really great things there. You know, the one of the big differences and you mentioned uh you know,
07:01
the hottest acronym right now Rag Retrieval gen uh retrieval augmented Generative A I, you know, the retrieval part is so important for that. And you know, when you think of a, a modern A I model, whether it's a pre trained model or something that you built yourself, the data that that model was built on is frozen in time on that day. And so, you know, a very simple example is,
07:23
well, what's the weather gonna be like tomorrow? That model has no idea how to do that. But with a Rag model, it knows where to go get that data source and something that's super important in the enterprises, you tell it where that trusted data source is. So the weather example is super easy. You know, my trusted data source is gonna be, you know, an online weather service.
07:44
But when you're asking a question of, you know, how do I uh you know, set up my, you know, latest pure storage appliance? Well, you want that trusted data source to be going to pure, you don't want it to just be going to, you know, somebody's blog on the internet, you know, who may, you know, happen to have the right answer or the wrong answer.
08:03
And so you know, the retrieval part of rag, first of all, you need your storage, your data very close to your model. So, you know, when customers are interacting with A I models, we call that inference. But that model running has to be very close to the data because the model could give you the
08:22
first part of the answer very quickly. But then if it takes a long time to actually get to the data store, parse the data and get the customer the answer back, it feels unnatural. And when we start talking about response times, you know, for a, you know, a complex A I model, you know, each word that's coming out has to be, you know, in the less than 100 millisecond range,
08:45
ideally less than 50 milliseconds. And as your model gets more complex, you need to get a bunch of different data because you may in that rag model need an answer. So it goes to a, you know, an internal company document, the model reads that document and realizes it needs to go look at another document that may be a PDF that's got an image on it.
09:07
You know, how do you process that? So you know, in not only having great infrastructure, you know, on the compute side and great networking, but having storage super close to those models, you know, enables that real time response to the customer and that natural response, which is what every enterprise wants out of their A I,
09:27
you know, in the early days of like phone response systems, what did everyone hit? You hit zero, you said operator, but great A I models right now, you actually want to interact with the A I model because it's so much faster to get you the right information than pushing zero and going to an operator. Now, you know, that's an old school concept.
09:44
But, you know, that's really what we're seeing with customers is is they're putting their, their I models in production. They've got to be very thoughtful about where their data lives, how close it is, how fresh it is. Um You know, and one of the, the newest concerns that we've started to see is data access and security policies. You know, you may have a rag that is for your
10:07
internal employees. Well, OK, you know, as, you know, a regular employee, I should be able to look at all my basic corporate information. But if I got the model, you're the CFO, you probably can look at other stuff. And we don't want, you know, the two, you don't want the, you know, the college intern to, to be able to access, you know,
10:25
confidential roadmap information on day one, you know, so data access and data security is super important in that space. Yeah, absolutely. And uh you know, as you were talking through, I think the responsiveness is a big piece we're seeing as well. Um You know, to use your previous example. Uh Hey,
10:42
what's the weather look like? Uh tomorrow if I have to wait for every word to come out? Only to realize that, hey, I asked the wrong question and I really meant to ask, what's the weather look like tomorrow in New York? Because I'm flying there again, that has, that has very real implications as to uh you know, how, how usable uh the technology is uh in, in kind of the business process.
11:06
And so we're seeing a lot of focus on um you know, not just, not just performance of the one system, but hey, to your point, uh where are the data sources that need to be fed into this? And um and what is the overall performance look like uh as a result of that? Um So, uh I, I guess you know, I guess maybe to close off,
11:29
um you know, curious if uh if you can share uh kind of insights from uh from your end and from a video in videos end, you know, if I'm an enterprise customer and I'm just getting started uh on my A I journey uh trying to figure out where do I jump in? Where do I head? Um What, what are your words of advice? What should I be doing? What should I be focusing on?
11:51
Yeah. You know, and uh you know, it, it's a great question and you know, something that I hear, you know, every time I'm at shows and conferences, you know, where do I get started? The first thing, you know, you've taken that first leap, you want to get started. Uh you know, so getting started with A I today
12:04
versus six months from now, get started today. Um You know, and one of the big things that we've done here at NVIDIA is started publishing a lot of uh online examples that you can interact with, you know, we call them labs, but you can see how a rag works. You can, you know, look at, you know, well, if I add this little bit of data to it myself,
12:26
obviously not company confidential data, but you know, it, you know, how does this process work? What does a demo look like? Because a big part of A I is inspiring people inside of your company, you know, to say, look what this can do. But then to go from, I've seen this great example, I'm excited about it.
12:45
You know, the real thing, you know, your next step is how do I interact with my data? Uh you know, and that's when you know, you, you and engage your NVIDIA team, engage your pure team. You know, the, the pure team knows how your data is laid out, knows where your storage is. And then together we can work on what's the first project because you don't want to try to,
13:06
you know, build 100 A I projects at once. You want to get one good one started. And as soon as people in your company start to see that project, everyone gets super excited and every department says, well, I want some of that, I want some of that. Um you know, but getting that first one up and running and don't make it too complicated.
13:23
You know, there's tons of things that you can do. You can pre trainin, you can guard rail, you know, all those aspects, you know, make it usable, make it safe. You know, that example of, you know, you don't want to let A I loose on all of your internal corporate information for anyone to see, you know, I've seen examples of that and it's gone sideways.
13:42
But like, you know, an example we use all the time uh with companies is like, do your first example on all of your public company news and then you can easily ask that question because there's no safety issue there because it's already fully public. It's on your website. But, you know, if you're wanting to look, you know, through your company's last five years of
14:04
press releases, you're like, I know we talked about this at some point. You don't wanna, you know, read two hours of text. That's a super easy example. And, you know, think of something like that and then that inspires people to say, what's the next most useful thing? And that's a big part of A I today.
14:21
It's not what's necessarily the thing that's gonna make you the most money the next day. But what's the most useful thing to your employees, to your customers? And then over time you're gonna figure out how to optimize and make more money and be more efficient. Yeah, I agree. You know, when I talk to customers, um uh you know, I very common to hear,
14:40
I've got 50 you know, 50 project ideas coming at me, left, right and center. Uh what do I like? Where do I get started? Uh And I always recommend, yeah, like you said, identify where you're going to get the highest ro i the most quickly. This is not a space where you wanna kick off a
14:56
two year deal development project before you start showing results. Um But then also uh because it's fast moving because you are gonna get uh you know, all of these different uh project ideas and initiatives uh planning for agility, right? And, and hey, what does that mean in terms of uh you know, what does that mean in terms of my application infrastructure,
15:15
in terms of I prepared for containers? Do I really understand uh you know, my data workflows? Um you know, but this, I think this idea of identifying where you can quickly show value and then expanding from there. Um While at the same time uh reviewing your data workflows, your data security posture, your data pro um I, I think from,
15:36
you know, from the customers, I'm chatting with, I, I think those are the uh the areas that again, the most successful uh clients we have are focused on uh more, uh I would say early on in the process. Um Well, Charlie, uh look, you know, I, I want to thank you a lot for the discussion today. Uh You know, uh uh very uh very media and a lot of topics we hit um for those of you
16:01
tuning in. Uh You know, I hope you enjoyed this episode of uh the Fireside chat series. Um Again, Charlie, thank you for joining us. Uh And uh I hope the audience, uh I hope you will join us for more chats in the series. Uh But until then, uh see you next time and I hope everyone has a great day.
16:19
Thanks. Everyone.
  • Video
  • NVIDIA

In this Fireside Chat, Pure Storage CTO Rob Lee welcomes Charlie Boyle, VP and General Manager of DGX Systems at NVIDIA, for an enlightening discussion on the current state and future trajectory of Artificial Intelligence (AI) in the business world.

Celebrating eight years since the launch of DGX, Boyle reflects on the AI industry's remarkable expansion from a niche research field to a ubiquitous enterprise solution. The conversation covers the evolution of AI accessibility, highlighting the shift from requiring a PhD for AI deployment to the democratization brought about by tools like ChatGPT and pre-trained AI models and emphasizing the importance of integrating unique enterprise data with these models to customize and enhance business applications.

We'll also delve into the practical aspects of AI implementation, including the significance of data proximity to AI models for real-time responsiveness and the role of data access and security policies in enterprise environments. We conclude with actionable advice for businesses embarking on their AI journey, stressing the importance of starting with manageable projects to demonstrate value and inspire organizational adoption.

09/2024
Seven Key Storage Considerations for Digital Pathology
Explore 7 key considerations your organization should consider when choosing data storage for digital pathology.
White Paper
4 pages
Continue Watching
We hope you found this preview valuable. To continue watching this video please provide your information below.
This site is protected by reCAPTCHA and the Google Privacy Policy and Terms of Service apply.
Your Browser Is No Longer Supported!

Older browsers often represent security risks. In order to deliver the best possible experience when using our site, please update to any of these latest browsers.