 Good afternoon everyone, welcome back to theCUBE. We're live at Dell Technologies World 23 at Mandalay Bay in Las Vegas. Lisa Martin with Dave Vellante. There's been a lot of news in just the last, I would say what, 26 hours or so, Dave. As part of our industry. It really is. More great news came out this morning from Dell and Nvidia. You're going to want to hear this if you're interested in generative AI. And who is it? One of our alumni is back with us, Veroen Chabra, SVP Product Marketing at Dell Technologies. And Carrie Brisky joins us as well. VP AI Software Product Management at Nvidia. It's great to have you both. Thank you for joining us. Thank you for having us. Yep, thanks for having us. Talk a little bit about, generative AI is the hottest topic. And we were just up here talking with Chuck Whitton and his kids are talking about it. Everyone's talking about it. It's got massive potential to change the future of the world, the future of enterprise IT. Talk about some of that potential, but some of the challenges that customers are coming to Dell and Nvidia to solve. Carrie, we'll start with you. Yeah, so I mean, obviously we've all felt the buzz about generative AI. I think what's exciting is that a few of us have been working on generative AI for actually a couple years. And so I'm glad that we've all had that aha moment, the iPhone moment that just made sense to how can we integrate it into business applications. The thing is, is that I mentioned earlier today that not every enterprise can train a foundational model from scratch. I mean, it's mountains of data you have to curate, a large amount of compute, the algorithmic resources that you have to know to how to scale across thousands of GPUs, all these algorithms. So what we've done today with Project Helix is provide these pre-trained foundational models and various sizes with our NVIDIA Nemo framework. And then, coupled with Dell, provide that to enterprises so you can get just a jumpstart on solving your business applications. And what's been the feedback from customers on Project Helix? Obviously announced this morning, Vareen, what are some of the things that you're hearing on the show floor? Yeah, Lisa, I think what we're finding as we talk to customers about generative AI in general is that as you would imagine with anything that's kind of hard to imagine, but still early stage, given how much hype there is about it, is that there is a spectrum of needs and people are on a journey here, right? So really the goal with Helix is to provide customers with help with generative AI enterprises wherever they are on their journey. If there are most sophisticated customers who want to build a model from scratch, a generative model from scratch, we want to help them with that. If they're taking a foundational model that's included in the solution from NVIDIA and they want to tune it with their own proprietary data, we want to help them with that. Let's say they figure all of that stuff out, they've got their application ready, they've got their models ready, they want to deploy this at scale, maybe for an internal back office process or transforming their customer experience, we want to help them with that as well. What is the training infrastructure? What is the inferencing infrastructure? How do you think about bringing all the components from Dell and NVIDIA together? It's just a great, great place to get started. How did Project Helix start? Take us back, when, how did you guys get together on that? I can take a stab at it and carry, you can chime in. Look, it may seem like, oh, this is a flash in the pan, but really the work on Project Helix started with the work that we're doing on the 16G PowerEd servers, right? So we have the XC9680 or the XC series in general, which are really built for AI workloads. There was a close partnership with NVIDIA to design an eight-way server that has eight GPUs in it and not just, you know, not just slap the GPUs in there. How do you tune the performance, the networking, even cooling requirements, et cetera to make sure that this can operate at scale? This work started, but like two years ago, if not more. And I think as we've been on that journey, we announced this at GTC this year, it's just the reaction has been so unbelievable, but it's also been very clear that we've got to do more for our customers. You know, customers are concerned about data privacy, data security, right? It could be their own privacy, their own privacy regulations or local privacy regulations, industry privacy regulations. Well, how do I get started? Where do I get started? How do I get the expertise needed to bring these together? So that's really what Helix is doing. It's taking the work we've already done together and really packaging that up so that we reduce the friction for our customers jointly. So if it weren't for ChatGPT, would you have announced it Dell Tech World? Part A, Part B is, would anybody have paid attention? So, ChatGPT, I don't know what that is, but I'm not familiar with that. I mean, I think it's smart that you have to have it as the year of AI, but you think you would have announced? Yeah, I mean, I think that, look, the work for 9680 has been happening for a long time. I think what's unique about Helix is we're taking the needs we're hearing from the customers because, you know, ChatGPT has made it front and center for our customers. I think that's the reason why we're calling it Project because we want to make sure we incorporate that feedback in there as well. The 9680 was going to happen regardless. Like we know that was going to happen. And AI workloads have only been growing, right? So since computer vision, you have the ImageNet moment, you've had the BERT moment, superhuman tests on the glue scores for natural language processing. So it's just as, then we had recommendation engines for neural recommendation engines and now generative AI. And actually, generative AI has been around for a couple of years as well too. It just was that flip moment with that everyone understood it. But, you know, we've had generative speech, we've had large language models, we've had large language models. So I think that, yes, of course, we would have announced it anyway. And with the NEMO framework, NEMO framework has been around for a couple of years. So we actually accelerate and help train it and for every single type of AI workload. But generative AI is just really important right now. The ChatGPT hype has been a gift because I think a lot more people pay attention. So we were talking earlier, Carrie, about I watched that interview between Jensen and Iliad. Is it Iliad? Amazing, you should watch it. He's the co-founder of OpenAI. And he was talking about how completion was the magic sauce of what they developed. And you were explaining, yeah, that's cool, but there's a lot of other, can you explain that and how it's different from what Project Helix is? Well, so the completion is part of the model, right? So that's what you're generating that next step, that next character, that next protein, that next molecule, that next part of speech, a voice. So that's a completion API. So when you're talking about Chat and the reinforcement learning through human feedback, of when you provide an answer to somebody and you're able to hone and say, what type of conversation would you like to have? Would you like to talk more technical? Would you like to be able to give short, curt answers? Would you like me to elaborate? So all of that is fed back in through that reinforcement learning of human feedback. And that's what makes it so important to be chatting. And then there's also one other thing about the chain of thought. These large language models, the large of the model, it's been proven that they improve the perplexity scores of being able to keep that conversation going. What was the thought that you just had? How am I continuing that? How can I have this multi-turn conversation keep going? And that's the chat aspect of completion and generative. And just to add a little bit about Helix and how that ties into this, Helix is really about building a horizontal platform, right? And if customers want to deploy an LLM, they can do that. If they want to deploy a diffusion model for image generation, they can do that, right? Or anything in between. It could be, as Carrie was educating me, something around healthcare. It's meant to really enable the use cases and the innovation that we know is coming in the enterprise that will be domain-specific and cater towards very, very specific use cases. Talk a little bit about building a generative AI cluster, infrastructure, software, really difficult. Dell and NVIDIA taking out a lot of that complexity for organizations. One of the things Michael Dell said yesterday in his keynote was that for companies that haven't already, are not already working with AI, they're already behind. Is Project Helix cannot be an accelerant for businesses that are going, help guys, we've got to do more than dip our toe in the water? Yes, I think the thing to remember about Helix is that the infrastructure from Dell, the software from Dell, the NVIDIA software, the NVIDIA GPU accelerators are available to customers today. Helix is really about bringing all of those together, creating blueprints, tailoring infrastructure and software and tuning it specifically for use cases depending on where our customers are. So it's really built, Lisa, as you said, for accelerating the journey, providing the expertise. Carey, anything to add? Yeah, I would just say that Project Helix provides freedom, right, because a lot of enterprises being hold into clouds or cloud services and now when you're able to keep that privacy, data security on-prem, you get the choice and freedom is choice. So I want to follow up on that because yesterday I heard Satya, we listened to him talking about laptops, could be a target. Jensen today said, yeah, a lot of this stuff has to be done on-prem. So part of me said, is that just because they're at Dell's house and they're being nice or is there a shift going on where, and I have talked to a lot of HPC folks in the last week or so because ISC is happening in Hamburg and interviewed a bunch of folks beforehand and they're like running half a million cores on-prem. They have no choice. But is there a shift going on? Will AI perpetuate in your view a shift, not a repatriation, but maybe a rebalancing of where people put work? Vignan's on that? I mean, the way I think about it is, in some respects, GenAI is no different from a lot of other workloads that customers are deploying today. Depending on their specific needs, data prices, regulations, performance requirements, there's going to be a mix of infrastructure and where it's located, right? And it's not just about cloud versus data center, Dave. It's also about the edge, right? And how the edge plays into it. I mean, I think really it would be foolish to pretend otherwise. It's really going to be a broad spectrum and certain workloads will be predisposed to running in data centers. Certain portions of the workloads will be at the edge and I definitely think the cloud will play a role in that as well. I echo that. I mean, I think everyone's kind of gotten over that. I would never put my data in the cloud. People have and they will continue to do so. But again, I just meant about, it's freedom, freedom of choice of where you want to do. You don't have to sacrifice certain privacy concerns for one solution or the other. You can now have it wherever you want it. But to your point about the edge, I'm glad you brought that up because I've written about this, is that most of the AI or much of it anyway in the spending is modeling in the cloud. And we've said, I think we were too conservative in our predictions going to end a decade. It's going to be mostly AI inferencing at the edge. So compress that by seven years based on my prediction. And it's happening right in front of us. And a lot of it's going to be this too. I get it. It's not just going to be on big million dollar boxes. But if you look at the, we're already there. Like if you look at form factors that we have in our PowerEdge line and VxRail storage, we've got a lot of form factors that don't cost anywhere near that, right? I mean, if you think about things like the XR4000 as Gil was talking about it today, these are much smaller form factors for things that have traditionally been much larger. Because there's going to be that interplay between whether it's edge and data center and edge and cloud and edge and both that allows you to offload some of that stuff there. But it also does mean a huge preponderance. And almost like, I think you said freedom, I would say democratization of this wherever you want to deploy it. And you know, Jensen touched on that. Again, I love this talk. He said, the thing is the app is miraculous. As an API, I can connect to anything. And it's a universal programming language called human. I mean, how powerful is that? And we can all touch in and use the term we talked about this. The data center is going to be called an AI factory. Like wow, okay. Love that. That is, I think it feels certainly like we're entering a new era and it's going to have impacts on productivity. Everybody wants to talk about job loss. You can't protect the past from the future. But what are you hearing from customers in terms of, do you think that this is going to bring productivity boom, like maybe not as much as the industrial evolution, but like we've never seen in the modern sort of computing era? What do you guys think about that? I would say absolutely no doubt. I mean, even if we're talking about the ways, Brune and I were talking about the ways we use gender of AI internally at each of our companies and just even for, say, IT service tickets, each of our IT service attendants is saving seven minutes for every single ticket that we, that it per day. So that's a lot of tickets per day. It's a lot of minutes saved. And so that's just one productivity gain, like one small drop in an ocean, right, of the amount of productivity gains that we can get for all the different use cases for gender of AI. And I think just to add, I think productivity is absolutely a place to go to and we will see a lot of initial benefits there. But one of the things that I think about is if you, as Jen felt was talking about today, as you open up these two, not just developers, these language models to rank and file employees, you know, people can now become coders when they couldn't do that, right? So maybe some employee in an enterprise had an idea about some really cool thing, but they were limited by their ability to code. Well, that creativity, that business acumen, is now unleashed, right? So if you think about the productivity is one angle, but the creativity and new outcomes, and it's a little bit of a cliche, but it's really hard to imagine for us where this is going to go. It just feels, as you said, we're on the precipice of something really, really big. Well, on that coding, because the tongue-in-cheek meme on Twitter is, time it takes me to write the code, like instant, and the time it takes me to debug it. But how real is that today? I mean, I have no doubt. It's like, this is like the first GUI we ever saw. It's like dial-up. You know, five years from now, we're going to be like, wow, we're going to be blown away today. It's going to just be, phew. But how real is that today? Are developers today using AI and is it making them significantly more productive? Absolutely, you have text to code, code to text, code completion. So it's happening today. I have a friend who's in a startup who has his own startup. He used to spend all his time on Substack. He said he just uses a chat GPD premium subscription now and you know, it's like. So he doesn't have to hunt and peck for the right syntax and the right code fragment. Yeah, yeah, and we're doing that internally at Dell as Jen was showing you today. We're opening up these models for our own developers and we certainly think that that's going to happen in other places as well. But to that point of hunting and pecking, you don't want one model to rule them all. That's why it's so important that every enterprise needs their own generative model because that one model is just a one view of the world. And so you need your own purview or aperture on the world and you be specialized in you and your data and your business because it might be, it's just a general knowledge of one thing or a little bit of everything but just not a subject matter expert in one thing. And I think that even as Jen was showing today in the keynote, the marketplace that we internally have for developers, it has open AI options, it has on-prem LLMs, you could do like Lama, you could do foundational models from NVIDIA. It is going to be kind of a multi-model world, I guess not modal, but multi-model world as well. I have three on my laptop. I have VARD, I have ChatGPT, which I pay for, and then I have theCUBE AI which we're trying to perfect and I compare the answers. A lot of times theCUBE AI answers are better. And I can get, what Varun said, I can get a clip of it. And it's because you have your data, right? Tune for that versus the other ones that you're talking about are general purpose models which will be good for certain use cases but the use cases that you guys have, your own data is going to help deliver a much better result for you and your customers. Yeah, and in my head I'm like, okay, I don't want to leak my IP because somebody else is going to be able to do this. How can I add more, enrich my data set? I mean, it's just mind blowing. It's not just enriching your data set. I think that's a great segue into guardrails, right? Being able to provide guardrails for LLMs, you keep it topical on the topics that you care about, that it's safety, it's not giving any toxic responses or unwanted answers, and then secure that it's not going off to some remote application and installing some malicious code. So being able to provide guardrails to large language models or to your CUBE app is going to be really, really important. Yeah, very cool. My last question for you guys is for every business that is looking to power the entire lifecycle of generative AI and blow our minds in the next year, the value prop, why Dell and NVIDIA? What do you guys say? Look, I think the biggest value prop, I would say, is the combination of the infrastructure and the software, you know, the accelerators and the expertise. I think the expertise is a really, really important place to start. You know, you'll be surprised at how many customers have, the basic questions that customers have, right? It's still early days. And I think sometimes the conversations are where do I start, right? So the expertise and the fact that this is not new stuff that, you know, we've been working on this for years, I think that combined knowledge and combined expertise, I would say is the biggest differentiator, but can I, from your point of view? Yeah, just being able to have a great partner, they take it to enterprises. I mean, NVIDIA is a small company. We need a partner like Dell to really help just bring our platform and democratize AI, like Varun said. Awesome guys, thank you so much for joining Dave and me. I cannot wait to see the direction project he looks goes in by next year. This has been fantastic. Thank you so much for your time. Thank you for having us. All right, our pleasure. For our guests and Dave Vellante, I'm Lisa Martin. Up next, Caitlyn Gordon is back with us with Maki Kapoor. They're going to be talking all things Apex, what the future looks like, ground to cloud, cloud to ground, air traffic control, you name it. Stick around, they'll be right here.