← Reshaping Workflows with Dell Pro Precision and NVIDIA RTX PRO GPUs
Reshaping Workflows with Dell Pro Precision and NVIDIA RTX PRO GPUs
Reshaping Broadcast Workflows with AI
00:38:42
AI is reshaping broadcast workflows. Learn how Dell, NVIDIA, and Qvest are powering real-world use cases from video search to personalization.
Loading summary
Transcript36 lines
- [00:05]
A
Welcome to Reshaping Workflows with Dell Pro Max PCs and Nvidia where innovation meets real world impact in high performance computing.
- [00:20]
B
Welcome back to another episode of Reshaping Workflows on Dell Pro Max and Nvidia RTX GPUs. I am Logan Lawler. Today we have a fantastic episode for a couple reasons. One, we have recorded a lot of episodes that weren't necessarily timed with events. Well, this one actually was purposely recorded to be able to release the week before IBC started so you can download it on your way to ibc. And two, we've covered a lot of topics from, you know, M and E, engineering, etc. Today we get to talk about broadcast specifically. Huge topic that happens at ibc. But that's not why I'm here. I'm here to guide the conversation. Let me introduce you to some of the experts you know with me today that are going to be guidance. So first off, Christoph, can you take a second tell the audience where you're from, who you work for and what you do?
- [01:06]
A
You Betcha. Christophe Ponsaert, 25 year experience in technology, media and entertainment industry, specifically in consulting as well. You know, obviously a very long background. It started off implementing one of the very first digital asset management systems that was enterprise wise for one of the major studios and kind of evolve from there. And you know, in terms of my focus it tends to be very kind of technology oriented. I'm one of the founders of Quest us. My major responsibilities with building out our entire engineering teams here in the US and over the last few years have transitioned all of my attention over to AI and helping to really evolve the shape of the AI usage within media entertainment companies from a Quest perspective to give a little background of who Quest is. You know, we are a consulting technology provider. We're the global leader in media entertainment. I think you'd be hard pressed to find a larger, deeper set of individuals in terms of background and experience in media entertainment. We are fully focused in that space. About 1200 folks globally, 300 here in the US and you know, we do everything from standing up entire broadcast center, figuring out the entire brick and mortar and the technology, the solutions and the implementation all the way through helping to bring out the streaming services and other things for the major studios and and broadcasters.
- [02:19]
B
Amazing. I had no Idea you had 1200 employees. I thought it was just you in the back room doing a bunch of coding. I guess I was wrong.
- [02:25]
A
Sometimes it feels like it Logan.
- [02:27]
B
Well, you know, sometimes it does. So Michael, can you do the Same. Tell everyone who you are, what your role is and where you work.
- [02:34]
C
Sure. Perfect. So. Thanks Logan. My name is Michael Kaplan. I'm a business development director at Nvidia. I've been here close to 15 years now. Prior to this I was at traditional broadcast vendors like Sony. I've been at graphics companies back in the day. But like Christoph had said, I think I've been in the industry maybe even a little longer than he has. But we both have been here for quite some time. So at Nvidia, I work in our media entertainment organization and what I do is sort of a segment sales partnership role to help drive AI innovation and solutions to the market. When you get to know Nvidia, we are really, really good with developers. We work with a lot of end users but sometimes they don't have the technical wherewithal or the teams that let them build solutions that we're offering and the tools that we offer. So we really lean on partnerships. Quest and Kristoff, we've had a chance to work together for quite some time. That's really the recipe for success. So when we focus on AI and we're going to talk about that today, I'm really geared towards that, trying to figure out what problems we can collectively put together, whether it's with a Quest or with Adele. We're really trying to line up our partnerships to take to market, to figure out the most challenging partnerships or I'm sorry, the most challenging problems that customers are facing and figure out what the right solution would be to drop in.
- [03:57]
B
I love that. And I'm glad you mentioned Dell as your favorite partner because I would have been a little, little hurt. A little hurt if you didn't. So you mentioned something, right? I think every, every company, whether it's in media, entertainment, broadcast engineering, are all facing what I'll call kind of the, the AI paralysis problem of hey, we know we need to do something, we're not sure exactly what to do. We're not sure, even less sure on how to actually do it. So you kind of teed up a great question and Christoph, I'll go to you specifically within kind of Broadcast and M and E. What do you think are. And I'll come to you Michael, after But what, what do you think some of the, the big problems that media entertainment companies broadcasts are facing today? Maybe like top couple of the ones that are just really stick out to you?
- [04:42]
A
Yeah, it's a great question and you're absolutely right. I think it is part of the challenge and it's, it's rightfully so. A challenge. Right. This has been a tidal wave of change of technology and opportunity. At first, it starts off as being, is this real? Will it really help? You know, these things are hallucinating. How much can I depend on it as an enterprise? I'm so used to things being predictable. Right. And you know, this technology as it kind of, as it supports you can also be unpredictable in the ways that, that it works, especially as it's kind of growing. So I think the. There's been a mounting amount of pressure from a leadership perspective to adopt these technologies because they're supposed to and they should be able to help you be more operationally efficient, help increase your revenues and all of these things. So, you know, step number one, and in terms of what we've noticed is we've seen a lot of these organizations try to first restructure how they're organized from a personnel perspective. Right. Because it's hard to make decisions when nobody's really the owner of said decision to move in particular direction. And so step one has been, and this has been an evolution probably over the last year, has been restructuring the organization to put yourself in a place where you have the right accountability and the right seats. And then those individuals can now start to set a vision, a guidelines and the policies around how do we move forward with AI and then start enabling budgets right behind that in order to push forward and build these things out. So that was kind of step one is a, actually helping these organizations figure out what should my organization look like in this new era of AI. And then, you know, step two ends up being where do I focus and start? Right. AI can be applied anywhere across the organization, which sounds like a great thing, but then it's like, I can't do everything everywhere. Right. It's not just about the technology, it's also about the process, the people, the change management that's involved. And so Media Entertainment's especially interesting as it relates to user adoption because you have gained guilds and talent and there's a lot of fear for job and work. And so in addition to trying to think about how can I more create operational efficiencies, it's also about how do I not impact these things? How do I adopt it in the right way? What we do a lot is help kind of on this assessment front of essentially looking at your, the technology strengths, your organizational value, what's going to bring organizational value then also user adoption. Right, Right. Because user adoption is kind of key to that recipe. And then we have an entire kind of proprietary process by which we can quickly go through a department or your whole organization with metrics supporting each one of those kind of three kind of dynamics.
- [07:15]
B
I love that. That's great. I want to quick follow. I think it's interesting that any real. Because you guys also, you know, you do implement consulting to all the way to implementation, right? So like it's funny that you talk about Org structure because I. That's not one I was expecting you to say honestly, because at Dell, not to give away what we're doing, but we do have kind of a centralized AI point at Dell where it is in charge of approving the tools use cases. They don't necessarily own the budget. The budget's owned by the individual businesses that are using the application. But they're looking at scale, they're looking at consolidation. Like why do we need two different chat agents, et cetera, et cetera. So I thought that was really, really fascinating. That was a great point. You know, Michael, going off what Kristoff said, is there anything else that you see as kind of challenges currently in the market from your perspective?
- [08:00]
C
Well, I, I agree with everything Christoph said and let me kind of put it into perspective from my point of view. I said I was at Nvidia for 15 years. Media today is so different than 15 years ago. I know that's like a no kidding. But what I mean by that, I'm going to kind of break it down into some of the biggest problems that I think this industry is facing. The first might be data analysis and insights. Right. So many of these customers are dealing with heaps of data and may not know what to do with it. So how do you extract meaningful insights from the data they have today to make sure that the experience a customer faces is personal? Right. I have adult children and their viewing habits are so different from mine, you got to be on point in minute one one or they're on to something else because there's so much out there. So I think personalization or hyper personalization and able to optimize what an experience can be is one of the major things that they need to do. And that leads into content creation and efficiency. You know, I do think there's so much out there, but it needs to be tuned for specific customers. And that whole creation flywheel you have to go through creating, you know, compelling entertainment. You've got to figure out what is the linear workflow, what's the online workflow, and how do you keep iterating at a faster speed in a somewhat personalized way.
- [09:24]
A
Right.
- [09:25]
C
And everything that I said, first statement, and the second statement leads into, it's about audience engagement. So audience engagement. I talked about personalization, but it could be an immersive experience. It could be at a venue, it could be on your phone. And everything has to be tailored from a distribution standpoint to get to the right clients at the right time. So it is a, it's a fascinating situation we're in with the problem statements, but content, we used to always say back in the day, content is king. I still think it is. I still think that's the same way. But maybe it's tailored, content is king. Or distributed perfectly, content is king. That's where I'm thinking about when I speak with customers. That's what kind of resonates.
- [10:04]
B
I agree. The personalization piece is so true. Right. And you know, I've had the privilege of not necessarily working with you, Michael, but others on the M and E Entertainment, M and E team at Nvidia. And I was shocked when I learned how much actual data science, like traditional data science, algorithmic based work goes into M and E. I honestly thought it was just people with cameras shooting out the desert, making a movie. And I know that sounds so stupid, but it's true. Like the amount of data that you're. That many companies are setting on for personalization. And I mean, I'll use an example, might not be relevant, but for example, the profile for my daughter on Netflix, very, very different from the profile that I have and rightfully so. Right. Because if she saw what I did, one, she'd probably be scarred forever. And two, it's like she wouldn't want to watch any of it. So personalization is absolutely huge. So Kristoff, I'll come back to you and kind of going in the, the personalization piece and the data science piece, what are some of either give us a couple of like use cases and then we'll ultimately get into kind of the solution on how you deploy and what toolkits and all that. But what are some of the other personalizations outside of a Netflix thing that you've seen in the market that you see companies starting to kind of move towards that quest and Nvidia could obviously help with.
- [11:18]
A
Well, and I'll touch on something that that relates to. I think everything you, you all just talked about is, you know, none of that was about dynamically creating brand new content creat using AI. Right. There's kind of the. It can be very easy to be distracted by AI's ability to generate new images, generate New video, completely new content that is not relevant to anything else going on. Everything we just talked about was really how can I streamline the workflow of data that I have today, getting access to more data and information that could be useful and then being able to make better decisions as to how I'm serving that data up to my end users. Right? And so I think that's one of, one of the interesting things about, you know, the landscape of AI and how it's come along is it can be very distracting to think of, oh, it's, it's all, it's all about kind of creating this new content. It's really about how can I better understand the content I'm creating today, the information around my entire landscape and then properly do the inferencing required to service up the information, you know, that is most relevant to my end user base. So, you know, I think we noticed some of this early on and we felt that there'd be a user adoption challenge as well on the content creation side. And so a lot of our focus over the last several years has actually been on using these transformer based models to extract data and information from existing content, right? From existing, you know, audio data, from existing images and then from existing catalog video. And then even most recently being able to do this on live video streams, right? Being able to extract meaningful information from what these models can see on these, on the various frames of a video, being able to do closed captioning, right? Language translation, a lot of things that help better personalize this content really comes down to I've got existing content, how can someone consume it better? Being able to transform the language, right? Into something that you can better, more appropriately understand so you can reason with the content is kind of key to being able to do that. So there's a lot of technology involved when you start to move up that type of content and layer, right? Taking text based data and using transformer based models on that to, to transform it is one thing. But when you start getting, and I'll go to the extreme of a live video broadcast now, you're talking about a lot of content, a lot of bits moving in all real time, right? You're taking the exact frames of things going on in a sports, a news, a broadcast of kind of event. You're, you want to be able to whole data, data streams of other known information about this event that's, you know, that's occurring right now. The people in the events, maybe the individuals inside, you know, the sport, sport event, taking other analytical information that Maybe you have in the background history of this event and this sport and what's happened before, knowing if this is the first time this has ever happened, right? And then you want to be able to inference on all of that in basically near real time to help highlight and bring up some of these anecdotes, either to an editor of the broadcast or being able to potentially start showing some of these anecdotes on the stream. And so being able to do that requires some dedicated hardware, right? It requires some dedicated. This is where your hardware, your GPUs, they all need to be connected in a way to support this very specific use case and tying that in. And so having that kind of this close partnership and relationship across Dell, you know, across Nvidia and across Quest, where we're kind of tying these use cases in to solve these particular problems. It's not just a new application problem, it's a hardware problem. And it's also even updating the engineering at the GPU level, which is why this collaboration also within Nvidia and being able to refine all of that in these ways is so important.
- [14:59]
B
I love that. I mean, and it's true like when you're, when you think about the data, right, and you think about recorded audio, you think recorded video, I mean, you're watching a live broadcast, let's say it's a baseball game, and you're able to real time, be able to caption and understand that video and basically be able to query against it to say, hey, has anyone ever hit four hits in a single inning? And then being able to reference that video from the archive and be like, oh yeah, you know, Pete Rose did it in 1976 or whatever, right? Like it is very powerful because without you're, you're furiously googling and you have no idea.
- [15:34]
A
And right now that is what, you know, from a broadcast perspective, right? They are also doing, it's a lot of manual activities put in place to do the research in real time, right? And the research in real time is on back catalogs that have really been tagged only manually at the video level. And maybe they broke it down into a manual clip, right? And then they've maybe manually tagged that. I mean, we're talking years and years and years of video back catalog footage that you can't access in this type of real time way. And so this is where being able to use, you know, video, use AI on video content to analyze the understanding of your entire back catalog, being able to find very specific clips on the spot in the moment can really unlock huge opportunities.
- [16:18]
B
I mean huge value. And like it comes back to kind of the personalization and the context of what you're providing. And I'll give you a quick example, Michael, I'm going to jump to you with a question is that with this podcast, right, I'm going to tell a somewhat embarrassing story is I am totally comfortable getting on camera and talking to people. That's my jam. I can do it all day. However, when it comes to writing and you two have received emails from me, not the greatest thing in the world. And I was asked, hey, this is really great content that you're doing, Logan. However, we would love to see it as a blog on Dell.com and I got cold sweats because I was like, oh God no. I don't want to have to listen to myself talk. I don't want to have to type out notes. So I basically on my Dell Pro Max, very simplistic thing I'm going to tell you, not enterprise scale. This is just for Logan, but was able to use an Nvidia parakeet model and Nemo and Nemo retriever to basically run the audio through caption it accurately putting it into a rag database and then queried against it asking it to write a blog post. Now was it perfect, was it perfect production to go on Dell.com Absolutely not. But was it 95% of the way there? So think about that in the context of broadcast where you have a nightly news report, you could quickly do that. I mean it used to take me hour, I'm like an hour, hour and 15 minutes to do that. But now I can do it once I have the audio in about five minutes. So think about that in a real time perspective. Right? So that kind of leads me into my question. You Michael, is that, you know, we kind of talked about, you know, it's really understanding the data, the captioning of the data, the context of the data and then being able to the goal in real time to it. So from an Nvidia standpoint, obviously we know Nvidia is more than just a GPU company. What are some of the softwares and SDKs that enable that type of real time interaction and contextualization of video?
- [18:05]
C
I'm going to absolutely get into that specifically and I'm going to start by if you listen to what Christoph said, he kind of got fairly specific in terms of some of the problem statements. We started off in a big picture situation, but the bottom line is, you know, solving the problems is not just one solution. And he Actually referenced a few different examples. You brought up baseball yourself. Right. So when you take a look at what Nvidia has, what Dell has, we kind of go from a computation and a software standpoint. Dell is helping us build the infrastructure and Quest is kind of putting everything together. So Nvidia with what Quest is taking advantage of is we've got a whole suite of software tools. Last year at a GTC in 2024, we introduced something called the Nvidia Inference Microservice or a nim. And a NIM is sort of a container based executable that could be done for language, image recognition, things like biology that aren't, you know, relative to the media space, but so many things. We took that a little bit further and we introduced, by putting together various NIMs in an agentic AI workflow pipeline, we created something we call a blueprint. One of the biggest blueprints that was interesting to the media space was something called video search and summarization. And that's the ability to take archive footage or real time live video inputs and start to understand what's happening in the content, start to tag it a little bit better. And again, you know, from an Nvidia standpoint, we have all these tools. Some of our third party ISVs take advantage. So we can go to market with the Quest and adele. Whether you want to build something and have Quest do it for you, or whether you want them to integrate a solution that we're working with at the developer level because that might be a piece of the puzzle that's easy to deploy. But we love, we love our positioning, we like the NIM reference architectures, we like what we're doing with blueprints. And by really aligning with the Quest, with Dell as the underpinnings to that infrastructure, it's a solid play for end users. And I think, you know, so many possibilities to go in it. And even some of the things that I've seen Quest doing from a solution standpoint, super relevant, super exciting. And you know, we're really, we're doing everything we can to make an impact, to figure out some of the problems that we talked about earlier.
- [20:33]
B
Kristoff, go ahead. And then I, I got something.
- [20:35]
A
Yeah, no, I was just going to add, I mean, I, what I love doing is calling kind of these accelerators right for our clients. And you know, everything that Nvidia has kind of built here are massive accelerators where we can accelerate our timeline and delivery of a particular solution to our clients. You know, because adoption has been fairly slow within media entertainment, we found ourselves a few years back being like we're not going to wait, right for our clients to start thinking about these problems themselves and engaging with us to build them. We're going to start building accelerators on top of what Nvidia is already doing so that we come to the table being able to show here's the really great example of it working right, the technology. This is not pie in the sky idea and maybe this will work. Let's prove that it can work. Let's, let's build that on top of other accelerators that are existing where energy and engineering is going into it. And then now we're coming into the marketplace with solutions where we can start deploying them in four weeks. Right. It's like we have accelerators across contract data extraction, which we started several years ago around image based extraction, video based extraction, live video extraction. So as well as kind of agentic AI frameworks. And so being able to piece these building blocks into real world examples with their data in a four week timeframe and just essentially explodes them into a path of going from POC into production. And I think we all know nothing really matters until something gets in production because that's where it really starts adding value.
- [22:02]
B
Absolutely agree. You made a good point. We're going to come back to it on nims. And this is the analogy I use on NIMS is NIMS are great because one, you can go to build.Nvidia.com youm can pull down a NIM and you can test it for free until you roll it in production. Then obviously there's a cost because you know Nvidia is not nonprofit but is the way to think of a NIM is think of and most people that are listening this, maybe you're a developer, maybe you're not, but it's not as just as simple as pulling down a model. Yeah, you can do that but then it's the updating, it's the quantitization, it's the optimization to run on the GPU which by bringing in a NIM in the container is that is done for you. That is done by Nvidia. That's why it is a service. So it gives you the ability to said hey, instead of going hiring 10 data scientists, let's just let Nvidia do it. Do what they do best. Right. And you, you midi, you mentioned video search and summarization. So I don't have it. We'll link it in the description below. But we have the A, we have the Dell Technologies booth at ibc obviously. But we also have the tech zone where I think Michael will be. I know Christoph will definitely be. Well, I'm setting it up. Well, we are actually running that demo on the GB10, which this is just a mock up unit, but I've actually got one upstairs connected to Nvidia right now downloading all the VSS stuff anyways. So fun. But yes, you'll please come by the booth, see it in action. You'll be able to win a cool prize, interact with it. Now granted you're going to see it on a GB10 scale, not a, you know, an AI factory scale. That takes some, that's taken into consideration, but you'll be able to see play kind of understand. So Christoph, you mentioned a really good point that the proof is in the pudding when it comes to deployment for any AI solution. We can kind of wax lyrically about how great it is, but really at the end of the day we're all in business, we're trying to find kind of the roi. Can you talk through some of the implementations you've done? You don't have to share any trade secrets or proprietary information, but what do you see is kind of once you do deploy the, you know, the joint Dell Nvidia Quest solution, what are some of the outcomes that they receive? But then what is kind of the ROI or you know, business value that they get back out of a solution you implement?
- [24:07]
A
Maybe I'll give a few examples here. One of our clients and what's interesting, I think about these models, capabilities, you know, working with GPUs as a whole is you're in a place where you're getting significant value already by this compiled model, right? It's got this knowledge, it has this ability to reason. And then what you have is you have a whole lot of startups that are, you know, in the world creating these various use cases and trying to sell you on these services. So there's a lot of pressure I think internally across organizations now as well, where they're being shown all these various capabilities of how, you know, these things can make their lives easier and they're now considering these things as tools and options. When you start thinking about like the media organizations and we've talked about how much content these organizations have, you know, do you want to start pushing this content to a third party SaaS provider that's built up their technology stack over the six last six, nine months through the backings of this extremely powerful model, or do you want to put some engineering time on your Own, leveraging that same capability to, to deploy lots of value for your organization where you can then own and your own roadmap, right? And I think that's really important and key because they have very specific use cases and they don't want to be held back and wait on various. On one vendor, two vendors, five vendors that have to work together to get this thing deployed. You know, on the one front, we've helped one of these clients start building up their internal video analysis engine, right? How to, how do you get kind of your internal content through these models, giving them these summarizations, giving them true event detection across videos, and then helping them really market across their entire organization. The fact that they are building this internally in a centralized group and showing off the capabilities internally, you almost have to do an internal marketing campaign to some degree to get the rest of the organization excited so they don't go off into other third parties and they start leveraging you and your services there. What we see in terms of pattern, right, is you start off maybe in the cloud, leveraging cloud services in order to prove out a particular concept, right? So you can get off the ground quickly. But what you start to see is, especially as you're dealing with larger content and then you're dealing with your entire back catalog, thinking about this ROI becomes critical to enabling this use case, right? And so what you then want to do is you want to start optimizing your implementation to start making better use of that. So instead of a token based pet method of passing every single frame across a video, you can imagine it's a lot of tokens. You're rendering 20,000 closed captioning files coming out of this one video. It is a lot of processing. You start to move towards more dedicated model, right? You start to move towards more specialized large language or video language models that are more efficient at working as well. And we've seen, you know, 60% cost savings and starting to come off of, you know, that, let's say cloud tokenized instance and going more toward a dedicated gpu, right, that you can own, that you can manage, that you can deploy as well as kind of managing smaller models to do that. And I think that can go all the way into kind of more, you know, local deployments as well in terms of trying to optimize for that. So that's kind of one scenario in which, you know, you may start off in the cloud for some of these POCs to get, you know, quick valuable. When you now you want to start to scale, you, you need to start Thinking about where should this live, how much of this should we own, how do we deploy this ourselves? And that's where you kind of go and drive some of that. Another interesting example, you know, somewhat different scope, instead of being video is more image based, right? We had a client approach us with some a use case where they create kind of their marketing image based content, right? Multi layered files that, that essentially they go, they take one hero shot of a particular film as an example and they'll create hundreds of versions of that exact, exact same hero image for all the different social media networks, right? But it's really the same multi layer image that is essentially taken its various layers moved in different shapes and sizes, you know, used to manage these various safe zones. And so one of the things they did is that there's gotta be an easier way to do this. And so we came back and said, hey, this is what we would do, right? We put an agentic AI framework together that would enable that entire process to be done using AI agents. It would break down this multi layered file, it would use a vision language model to look at every single layer, decide what is this layer? Is this the characters, the cast, the crew, you know, is it the background? And then it would be able to in an agentic way, figure out a plan to move these layers around into the appropriate orientation and deliver that solution. And so within a four week timeframe, leveraging these building blocks right around agentic workflows and so forth, able to deliver a solution which walked through that entire workflow and started generating multiple versions on dynamics, safety zones and the rest of the. And it kind of blew their mind. And so what they started to do then was started to again try to democratize this across the organization, start building this in a more scalable way so that they now have a more agentic workforce that can do these image based validations and know if a head's cropped or Texas cropped and remove and shape these layers into the right orientation. So there's a lot of different ways to approach this problem, but it's all about trying to find ways in which to pulling this data together into this workflow and bringing them closer to the models as possible.
- [29:10]
B
I love that. This is fantastic. So, you know, I think you made a really good point and I'll pass it to you Michael, here in just a second and then we're getting kind of up on time. But really think about this. You mentioned kind of the data flywheel, right? This is more of like the AI factory flywheel, right? Where you have the people for the implementation, which to be honest, we don't have an army of people that know M and E that are developers that have industry experience. So that's not us, that's Quest. We're not making the GPUs and the SDKs, but we are making the underlying hardware, whether it be Delpromax Workstations or PowerEdge 9680 servers. So you know, Michael, you kind of got a view of kind of the entire the Dell AI factory with Nvidia landscape. You know, is there anything that you would add specifically from any deployment examples or anything with a customer who's kind of came in and been able to optimize within the Dell AI factory that where you've been able to been like, wow, like we absolutely crushed this. This was absolutely life changing for this customer or business changing for this customer.
- [30:07]
C
Yeah, I do like what Dell's doing. You know, you guys have a very pliable framework, meaning you could start small. And you referenced and named the workstations or the servers. So with clients who may have a workstation already or might be looking to acquire new ones, it's a great step one, it's a great chance to test of poc. There's, you know, models that can fit into the memory footprint of some of our new GPUs that recently were introduced, part of the Blackwell architecture. And then when you have to scale, you have the server suite that also takes advantage of the latest and greatest Nvidia tech that can do that for you. So I like the way you approach the market. I like your flexibility in terms of start where you need to as a customer and then again from a networking and a CPU and GPU standpoint, you definitely have the foundation for a project to scale right at the end of the day, this whole talk has been really woven around three pillars. The Dell, the Nvidia and the Quest enablement. You talked earlier about the build.invideo.com website and you mentioned the NIM and we talked about blueprints. The benefit of the NIM that some customers don't know is we've done all the hard work for you. Right? Whatever the model might be, we will get our hands on it, we will optimize it, we will put security in it and then we containerize it to make it deployable. Same thing with the blueprints so that the end user customer is ready to go and a quest can come in and start to deploy. But the infrastructure and the platforms need to be there and, and the Dell AI factory is such a perfect step to doing that. And like I said before, you got three, three folks working this together for an end user. And I think together it's a very successful trilogy.
- [32:01]
B
I love it. Successful trilogy. I like it. I'm going to steal that, so I'm going to steal that. So what I like to do at the end of the episode is, Kristoff, take 30 seconds. Imagine someone just joined the episode. What is that 30 seconds that they need to walk away with and remember? Because, you know, you only remember the thing at the first and at the end, right? So what is that 30 seconds that they need to walk away? And we could probably do it about. You could probably talk about Quest, I'll do it about Dell and Michael, you can talk about Nvidia. So what do people need to walk away with 30 seconds remembering about quests in this episode?
- [32:36]
A
Well, I think one of the most important things is that we're working very closely together. And I think doing that, bringing either any one of us in, I think brings all of us in. And it's an accelerant to them being able to see value. Right. In a very real, tangible way where they can see results. At the end of the day, right? I think all of us are here because at the end of the day, it's really about the results. And there are many use cases, you know, across the industry, and we are so actively involved across, really the majority of leading broadcasters, trying to paint this picture of what the agentic AI future looks like. Right. And there's so much value, I think, in the expansion, the application, the deployment, the reusability, the efficiency that gets deployed as part of this. And it's not just a application story, it's a, it's a hardware story, it's an accelerating GPU story. And I think the more that you build your development environment, like your production environment, the easier that entire flow from, from, from development through production will be realized. I think the last thing, and we probably didn't touch on it too much, is, you know, a lot of these models started off very, very large, right. It started off kind of as you couldn't have done this without the cloud, right, to be able to do it. But what you're going to start seeing from a pattern perspective is these, these models are getting smaller, more efficient and effective for the purposes that they're in. And you can start seeing that with some of these most recent reasoning models that are coming out that are deployable on your workstation, deployable on your local device, and what that's doing is enabling you to hit a lot of these other use cases that hold privacy, security, even speed. Right? You get so many more efficiencies being able to run something right there where your content is and where your information is on your device in a lot of cases. And so it's a very exciting time. Really appreciate the opportunity to be working together and it's been so much fun to be successful.
- [34:30]
B
So so far, I love it. All right, Michael, same question to you.
- [34:34]
C
Perfect. So Nvidia, if people don't understand who we are, you know, we started as a graphics company and we've turned ourselves into an accelerated computing company. And graphics still comes along on that ride for media entertainment. We're in a perfect situation because we can take advantage of everything that Nvidia has done historically and recently from a computational and graphic standpoint. What we do is we try to do accelerated computing both graphically and computationally. We look at the flywheel as creation, distribution, and then analytics to see how it's been. Do you need to redo the content creation? Do you need to distribute it differently? And we build out solutions, software, NIMs, blueprints, libraries, SDKs that can help that entire circle or flywheel that an end user can take advantage of. More times than not, they may not be able to do things themselves and have to rely on the partnerships that we have with Quest to glue everything together. We're a big partner company, we're a big ecosystem company, and we're here to listen to what the challenges really are. And we will align with the right partners to figure out what the solution needs to be.
- [35:45]
B
I love it. Kristoff, tell everyone where they can find you on social media. And, you know, if you want to point them to anywhere on the Quest website, give that URL and we'll link it in the description below.
- [35:56]
A
I appreciate that you can essentially find us on quest.com that's Q V E S T. It may sound like Q W E S T but it is q v e dash s dash t com and you know, most prominently I'm on LinkedIn. Christophe Ponsaert under Quest. You can also find me on X. I'm trying to post more, but I believe I'm under C. Ponsaert.
- [36:17]
B
I love it. Michael, same question to you.
- [36:20]
C
Same thing. You know, obviously look me up on LinkedIn. Michael Kaplan with a K at Nvidia. I'm also on X. I think the only people to follow me might be my kids, but definitely come on and take a look. And then Nvidia.com like it's a broad website build.Nvidia.com from an AI standpoint, you want to test things, try things out and just do some existing tests to then pull in a quest and say, let's go. Please do that. So I'm out there. If you need me, you can find me.
- [36:49]
B
I love that. So, yeah, just to recap this episode, I think this is the first one that we've had where it wasn't just telling the Dell or the partner story or the Dell and the Nvidia story. This is where we kind of full circle, right? It's like the three best friends that anyone could have. If you know which movie I'm referencing right now. But you have the data, the industry knowledge from Quest able to bring that insight, those years of experience, those development resources to be able to tag, summarize your data using the Nvidia, you know, SDK search and summarization, blueprints and NIMs. Obviously GPUs. Can't forget to mention the GPUs all running on Dell hardware, which is, you know, death side if you want to pilot or you want to do a proof of concept all the way up to ultimately the data center with more GPUs than my 11 year old could probably count. So anything in between, we can meet you where you are in the journey. And the thing you need to remember is one. Come to the IBC Tech zone, try out the video search and summarization on the GB10. Or, well, let me get the branding right. The Dell Pro Max with GB10, try out the demo and with that all you got to do is take the first step. Quest, Nvidia and Dell Technologies are there for you. And with that, this is reshaping workflows with Dell Pro Max and Nvidia RTX GPUs. Until next time, keep all your video search and summarization workflows running locally on Dell Pro max and Dell PowerEdge servers and we'll see you on the next one.
- [38:31]
A
Do what you want. Do what you want.
- [38:34]
C
This podcast was produced in partnership with Amaze Media Labs.