Red Hat Summit 2024 | Jeni Barovian | Intel & Steven Huels | Red Hat

Clips
News
More from Red Hat Summit 2024

Jeni Barovian Panhorst

President & GM, Data Center AI Solutions Strategy & Product Management

Intel

Steven Huels

GM, AI Business Unit

Red Hat

Intel and Red Hat join forces to accelerate generative AI innovation and adoption

During Red Hat Summit in Denver this week, Red Hat Inc. announced plans to bring Intel Corp.’s most advanced artificial intelligence processors to its customers.By making cloud-hosted versions of Intel’s Gaudi AI accelerators, along with the chipmaker’s Ultra and Core central processing units, Xeon, and the Arc graphics processing unit available, Red Hat is acknowledging the role that major chip suppliers such as Intel are having in the model-driven world of AI

play_circle_outline The transformative power of AI and its impact on enterprises

play_circle_outline Intel's Gaudi accelerators and AI PCs

play_circle_outline Red Hat's hybrid cloud approach to AI

play_circle_outline Exploring Opportunities for Developers and Fostering Engagement through Intel's Open Platform for Enterprise AI

Info

Jeni Barovian | Intel & Steven Huels | Red Hat

Jeni Barovian Panhorst

President & GM, Data Center AI Solutions Strategy & Product Management Intel

Steven Huels

GM, AI Business Unit Red Hat

search

>> Hello and welcome back to day two of Red Hat Summit and AnsibleFest here in beautiful Denver. A little bit chilly today, but what's warming our hearts is all the openness and all the open source that's going on today. And we're digging in. I'm Rob Strechay, got Paul Gillan with me. Rebecca is flying around here taking a little break. And I think what's really fun has been, I think the openness and the ecosystem. Today has really been a lot about the ecosystem and embracing this. And really, we're getting to see a lot into those relationships.

Paul Gillin

>> Well, it's all about ecosystem at Red Hat because, of course everything is open source. So the business model is very dependent on partnerships.

>> Absolutely. And we're joined by Steven Huels, who is the VP and GM of the AI business unit at Red Hat. Welcome back. It was like you were just here.

Steven Huels

>> It was, thanks for having me back.

>> And we also have Jeni Barovian who works in the data center silicon platforms at Intel. Hopefully I didn't screw that up too badly.

Jeni Barovian Panhorst

>> You got it. Great to be here.

>> Okay, excellent.

Jeni Barovian Panhorst

>> Thanks for having me.

>> Obviously, Pat was on stage and-

Jeni Barovian Panhorst

>> Yesterday. Yeah, he was.

>> You can definitely see the relationship. And by the way, we had Matt Hicks on yesterday and we were talking about that, and just knowing how difficult that is for both people on both sides of that, because it was live.

Jeni Barovian Panhorst

>> It was live.

>> And it came off awesome. And I think, again, it's just a lot of fun. And I think what it also underscored is just the rate of change and innovation that is going on in this industry and how these partnerships and how you have to really tie really tightly together. So I think to both of you, take a minute and share your perspective on how this AI is really transformative and how it's really changing things. Why don't we start with you, Jeni?

Jeni Barovian Panhorst

>> Yeah, it's really interesting because the history of AI is really intertwined with the history of computing, honestly, all the way back to the beginning. And so a lot of people have pondered this question of, "Why now? Why is this the boom? Why is this the inflection point for a tremendous amount of value being derived from AI?" And I think it's really a couple of different things that are coming together. Obviously we've had this massive explosion in data that we've been trying to harness the power of over the course of the past couple of decades. And now what you have is the computational capability really catching up with the compute resources available in these machines and AI focused cloud data centers that are being built out. But really on top of that, it's the innovation that's happening. It's the model innovation to really start realizing the promise of deriving value from massive amounts of data. And really what we're faced with now is, "Okay, how enterprises specifically harness a lot of innovation that began in the cloud and all the excitement around generative AI and large language models, how do we apply that to an enterprise environment?" And so that's really where I think we're collectively focused between Intel and Red Hat is how do we address those enterprise needs. So when we look at what Intel brings to the table in terms of that end-to-end set of opportunities that enterprises are focusing on addressing, we've got a set of capabilities in the cloud, in data centers, with Intel Xeon processors with AI capabilities like advanced matrix extensions or Intel AMX. We've got our Intel Gaudi accelerators, which Pat talked about yesterday with Matt on stage. A lot of excitement around the acceleration capability that we're offering there. We've got solutions at the edge. We were actually just talking right before we started filming about how tremendous amount of innovation happening at the edge and a lot of different scale of computational power that's required there. And we've got our Intel Xeon processors, our core ultra processors. We've got our Arc GPUs as a way to drive that capability for inference at the edge. And then we also have our AI PCs to empower and enable developers to develop AI solutions more quickly. So we're really excited about all of these different assets that we bring for this explosion and build out of AI for enterprise needs.

>> Yeah, no. And Steven, again ...

Steven Huels

>> Yeah, the appetite for AI right now is insatiable. And I think what really changed was the accessibility of what's been out there. AI has existed for a long time, but with generative AI, it really captured everyone's imagination. And now the expectation is across entire enterprises, every line of business is going to interact with AI. You look at the demand that drives and the ideas now that are being generated because of the potential impacts, and it is like everything Jeni mentioned. The imagination for where workloads will be run, whether it's inference and training, everyone keeps pushing the boundaries. So it's exciting to be a part of.

Paul Gillin

>> Jeni, I want to ask you, you're facing a competitor in NVIDIA who has got a very simple story. It's all about the GPU, and really is in the seat right now. How is Intel coming at this market differently?

Jeni Barovian Panhorst

>> We know that the future of computing is accelerated computing, and we know that there is a need and an appetite from our customers and our partners to deliver alternatives in the market for accelerators and GPUs. And that's really what we're bringing to the table with Gaudi. And for those of you who had a chance to watch the keynote yesterday when Pat did that live interchange with Matt Hicks, he talked about the excitement that was generated, particularly around Gaudi 3. We actually announced this just about a month ago at Intel Vision in actually my hometown in Phoenix. And we talked about a lot of different things. We talked about the performance that it's bringing, really competitive and compelling performance. It is a third generation architecture, so by comparison to the prior generation of Gaudi, it's bringing 4X the computing capability, 2X the networking performance, 1.5X The memory bandwidth. And it is very competitive versus other solutions that are in market today in both raw performance and particularly performance per dollar in terms of being able to deliver cost-effectiveness. But it's also addressing specific needs in enterprise. So we're expanding the OEM ecosystem. All of the four major OEMs across Dell Technologies, HP, Supermicro, Lenovo, all announce that they would be in market with solutions with Gaudi 3 when we bring it to market later this year in the second half. And we're also expanding the product line as well to support a PCI Express form factor, which is really interesting to enterprises actually, because you can deploy the technology and a form factor that fits into standard servers. And so we know that that's critical, but we also know that there are opportunities to combine the Gaudi accelerator technology with Intel Xeon technology as well. And so when you look at, for example, a really critical workflow in enterprises in the area of RAGs, retrieval augmented generation, harnessing the power of LLMs, but applying these other sources of knowledge, we've demoed ... and actually over in our show floor right around the corner, we've demoed a capability of bringing together both Xeon and Gaudi to service different portions of that workflow. So we really look at it as the future is definitely accelerated compute, but we're bringing multiple different assets to the table.

Paul Gillin

>> What's the difference between an accelerator and a parallel GPU architecture?

Jeni Barovian Panhorst

>> So Gaudi is an accelerator. We also have GPUs in our portfolio as well, and we definitely have a forward-looking roadmap that focuses on GPUs as well. We are offering developers access to the accelerators that we have with open programming models, leveraging open frameworks like PyTorch and ensuring that they can bring their models that they're developing on GPUs today onto those accelerators.

Steven Huels

>> I think it adds a lot of value. So from a Red Hat perspective, we get demands from customers to run across a lot of different hardware footprints. And when you start talking about being able to run these generative AI models maybe in edge locations, not everything's going to require a GPU. There's a lot of technology that's driving these generative AI models to be smaller, more efficient, more effective. That opens up a lot of opportunities for smaller processors, lower power consumption, lower heat footprint, more efficient, and that's, I think, what enterprises are really looking for. But all of that across a consistent stack gives them optionality in how they build and then where they deploy.

>> Yeah, I think in fact, I was going to hit on that because it was funny. We were talking about how things get deployed out to the edge in particular when you're talking about the fact that you need ... latency is a real thing. And especially when you're doing certain data apps and having gen AI being part of a data app and maybe it's actually deployed out to a 5G cell tower, and you want to minimize that. You want to do the inference out at the edge and be able to be responsive as part of an app. Because I think longer term ... And in fact we had Mo on from Red Hat who's engineering head for InstructLab, we had her on prior to that at KubeCon just a few months ago and we were talking about UX and some of the different ways that people are going to interact with these gen AI products, and it's not always going to be through a prompt. But where are you seeing, or how are you seeing and suggesting to enterprises that they really take a strategic approach to gen AI, and what are some of the things that you're seeing and helping them with?

Steven Huels

>> So from a Red Hat perspective, the rate of change in innovation and AI is explosive right now. So companies know they're going to have to take some speculative bets on technologies that might not take off, but they don't want to compromise their entire AI investment. So what we talk to them a lot about is our hybrid cloud story, especially with OpenShift AI, where we can provide a core platform that allows you to build, train, tune, deploy, monitor, and manage those models, but still gives composability. So if you want flexibility to bring in some of those speculative bets, if it doesn't work out, you're not compromising your entire investment, you can still bring in another technology and augment that with even your own innovation and open source innovation. And then underpinning that is a lot of our ecosystem. So we give them the optionality based on the type of workload, the specific use case. We can work with different types of accelerators, we can work on CPU. As much as we talk about generative AI, the bulk of what's out there is still predictive AI, and we can also run those workloads together, so we can do predictive and generative. And so really, that's our core message to them is if you're betting on Red Hat, you're going to be able to have a stable platform for the future that can absorb the changes that are coming and still work with all the innovation that's coming out of the ecosystem.

Paul Gillin

>> Jeni, I'm curious, your title is Data Center AI Product Management. Does that imply that Intel believes that the AI training is going to move largely to the data centers, or is it going to be a largely cloud phenomenon?

Jeni Barovian Panhorst

>> Well, certainly there's training that's happening in cloud data centers, but we're seeing definitely enterprise interest to not just not only harness the capabilities of training in their cloud deployments with the hyperscalers, their hybrid cloud deployments, as Steven was saying, but also we're seeing an increased interest in on-prem deployments for both training and for inference, although inference certainly is dominant in on-prem deployments and certainly at the edge.

>> Yeah, I think again, as we look at it, there's also the whole idea that Intel really does lean into the open aspect of this as well. Help people understand some of the announcements. I think Pat made some announcements recently with the Linux Foundation. What is it that really, Intel is leaning in with open source and with developers?

Jeni Barovian Panhorst

>> Yeah, open source has been critical to certainly everything we've done in partnership with Red Hat over the course of the past 30-ish years or so, and certainly what we're driving now with our approach to AI. I mentioned before taking advantage of open frameworks like PyTorch, model communities like Hugging Face. You mentioned InstructLab. I think there's a lot of really exciting innovation that's going on there in terms of building a community approach to adapting the power of large language models to different enterprise domains. But I think we're alluding to the announcement that Pat made around OPEA, or the Open Platform for Enterprise AI, and we initially talked about that a bit at Intel Vision about a month ago. The community launched in Linux Foundation immediately thereafter. There's about 15 companies that are signed up at this point, including Red Hat. We're really excited to get this community off the ground. And again, it's really trying to take an open and community approach to accelerating that pace of innovation and pace of development so enterprises can really harness the power of gen AI. And so I mentioned before a little bit about RAGs. We're contributing our RAG implementation to OPEAI as one area to focus on kickstarting that. Things are moving very quickly. We're going to be hosting community days next week. So there's an opportunity for anyone to get in on the ground floor and participate as we continue to build out the technical framework and start to build up the working groups there. So definitely a good opportunity for engagement.

Steven Huels

>> And a lot of the work we're doing focuses a lot on the consumability of the integrations. At the end of the day, the thing that's going to make AI successful is customer's ability to consume it. While it's great to create it, and it's exciting to play with a lot of the models, really it's the ability to roll it into production that is going to generate the value. When we talk about the integration we've done with enterprise AI, it really is a full stack integration, where from a customer perspective, it's invisible to you. The components are just there, you're getting access to the optimizations, whether it's at the accelerator layer, at the runtime layer for the model, at the developer layer for the various libraries. So it's an integrated experience that helps you generate value and put those models into production.

Paul Gillin

>> You mentioned earlier that about 40% of PCs shipped this year are expected to be AI PCs. What exactly does that mean? And of course, the desktop is Intel's wheelhouse. How do you look at the future? How do PCs evolve as AI machines?

Jeni Barovian Panhorst

>> I think there's a couple of different ways to look at it. I think there's certainly applications that we're all using every day, if you think about, for example, video conferencing applications and the idea to even just make visual processing much smarter. How do we make things like background images and blur and other enhancements to the ways that we communicate on a daily basis? That's harnessing the capabilities of GPUs and enhanced AI capabilities on an ongoing basis. There certainly is an opportunity for smaller language models running locally on a PC for very low latency and high responsiveness just as part of our normal usage of assistants and copilots on an ongoing basis. And so a tremendous amount of innovation happening there just as part of productivity applications, developer applications. And then that's actually where I wanted to take it next, which is how do we enable developers to move faster? So actually harnessing the GPU capabilities that exist on the platform as a proxy for what they're developing for ultimately in an enterprise data center or in the cloud. So it's a way to get started on your development and build code that ultimately scales to a broader deployment.

Steven Huels

>> And we're seeing the same trend. You mentioned InstructLab, and I'm sure Matt mentioned RHEL AI and what we're delivering there. We are seeing a demand for developers to be able to work with these language models on their PCs and then be able to move those up into production. And so we're taking advantage of all the innovation that's happening on AI PCs. And then when you're ready to move RHEL AI into production, we can roll that into an OpenShift AI and roll it across the enterprise. So we're really starting to mirror what you saw for application development now for AI, and again, taking big steps toward overall consumability.

>> Yeah, no, I think that totally makes sense, and I think that's a great place to leave it, beyond the fact that how do people find out more? I hear there's a blog that they can sign up for, something like that.

Jeni Barovian Panhorst

>> Oh, there's all kinds of stuff. We've got an AI blog that you can sign up for. The other thing that I would encourage developers to check out in particular, we've talked a little bit about Gaudi, we've gotten a lot of questions asked about, "How do we get access to it?" One thing that Pat talked about yesterday as well that I'll emphasize again, is that we do have Gaudi 2, and will have Gaudi 3 instances on Intel Tiber Developer Cloud, empowered by Red Hat and Dell Technologies. And so that is another way for developers to get access and develop for now and into the future. Certainly there's an opportunity to also engage in various communities. OPEAI is one place. But like you said, this whole conversation is about open source and communities, and so lots of opportunities to develop the open approach to enterprise AI in the future.

Steven Huels

>> Yeah. And here, you don't have to go far. I think I saw four or five of the InstructLab sweatshirts wander by. So join the InstructLab community, get started with large language models on your laptop, and then have a look at us at OpenShift AI as well.

>> Not to mention Podman.

Steven Huels

>> Yeah, Podman AI.

>> I guess I'll help you out with that. But hey, thank you both for coming on board. Really appreciate it. You did wonderful, and hope to have you back soon.

Jeni Barovian Panhorst

>> All right. Thank you so much.

Steven Huels

>> Thanks for having me.

>> And thank you for watching this episode of theCUBE from Red Hat Summit and AnsibleFest here in sunny, frigid-ish Denver today. High altitude's getting to me, getting a little wacky. But we're still here. Got a lot more to come. So stay tuned to theCUBE, the leader in high-tech information, analysis and news.

Jeni Barovian | Intel & Steven Huels | Red Hat

search

>> Hello and welcome back to day two of Red Hat Summit and AnsibleFest here in beautiful Denver. A little bit chilly today, but what's warming our hearts is all the openness and all the open source that's going on today. And we're digging in. I'm Rob Strechay, got Paul Gillan with me. Rebecca is flying around here taking a little break. And I think what's really fun has been, I think the openness and the ecosystem. Today has really been a lot about the ecosystem and embracing this. And really, we're getting to see a lot into those relationships.

Paul Gillin

>> Well, it's all about ecosystem at Red Hat because, of course everything is open source. So the business model is very dependent on partnerships.

>> Absolutely. And we're joined by Steven Huels, who is the VP and GM of the AI business unit at Red Hat. Welcome back. It was like you were just here.

Steven Huels

>> It was, thanks for having me back.

>> And we also have Jeni Barovian who works in the data center silicon platforms at Intel. Hopefully I didn't screw that up too badly.

Jeni Barovian Panhorst

>> You got it. Great to be here.

>> Okay, excellent.

Jeni Barovian Panhorst

>> Thanks for having me.

>> Obviously, Pat was on stage and-

Jeni Barovian Panhorst

>> Yesterday. Yeah, he was.

>> You can definitely see the relationship. And by the way, we had Matt Hicks on yesterday and we were talking about that, and just knowing how difficult that is for both people on both sides of that, because it was live.

Jeni Barovian Panhorst

>> It was live.

>> And it came off awesome. And I think, again, it's just a lot of fun. And I think what it also underscored is just the rate of change and innovation that is going on in this industry and how these partnerships and how you have to really tie really tightly together. So I think to both of you, take a minute and share your perspective on how this AI is really transformative and how it's really changing things. Why don't we start with you, Jeni?

Jeni Barovian Panhorst

>> Yeah, it's really interesting because the history of AI is really intertwined with the history of computing, honestly, all the way back to the beginning. And so a lot of people have pondered this question of, "Why now? Why is this the boom? Why is this the inflection point for a tremendous amount of value being derived from AI?" And I think it's really a couple of different things that are coming together. Obviously we've had this massive explosion in data that we've been trying to harness the power of over the course of the past couple of decades. And now what you have is the computational capability really catching up with the compute resources available in these machines and AI focused cloud data centers that are being built out. But really on top of that, it's the innovation that's happening. It's the model innovation to really start realizing the promise of deriving value from massive amounts of data. And really what we're faced with now is, "Okay, how enterprises specifically harness a lot of innovation that began in the cloud and all the excitement around generative AI and large language models, how do we apply that to an enterprise environment?" And so that's really where I think we're collectively focused between Intel and Red Hat is how do we address those enterprise needs. So when we look at what Intel brings to the table in terms of that end-to-end set of opportunities that enterprises are focusing on addressing, we've got a set of capabilities in the cloud, in data centers, with Intel Xeon processors with AI capabilities like advanced matrix extensions or Intel AMX. We've got our Intel Gaudi accelerators, which Pat talked about yesterday with Matt on stage. A lot of excitement around the acceleration capability that we're offering there. We've got solutions at the edge. We were actually just talking right before we started filming about how tremendous amount of innovation happening at the edge and a lot of different scale of computational power that's required there. And we've got our Intel Xeon processors, our core ultra processors. We've got our Arc GPUs as a way to drive that capability for inference at the edge. And then we also have our AI PCs to empower and enable developers to develop AI solutions more quickly. So we're really excited about all of these different assets that we bring for this explosion and build out of AI for enterprise needs.

>> Yeah, no. And Steven, again ...

Steven Huels

>> Yeah, the appetite for AI right now is insatiable. And I think what really changed was the accessibility of what's been out there. AI has existed for a long time, but with generative AI, it really captured everyone's imagination. And now the expectation is across entire enterprises, every line of business is going to interact with AI. You look at the demand that drives and the ideas now that are being generated because of the potential impacts, and it is like everything Jeni mentioned. The imagination for where workloads will be run, whether it's inference and training, everyone keeps pushing the boundaries. So it's exciting to be a part of.

Paul Gillin

>> Jeni, I want to ask you, you're facing a competitor in NVIDIA who has got a very simple story. It's all about the GPU, and really is in the seat right now. How is Intel coming at this market differently?

Jeni Barovian Panhorst

>> We know that the future of computing is accelerated computing, and we know that there is a need and an appetite from our customers and our partners to deliver alternatives in the market for accelerators and GPUs. And that's really what we're bringing to the table with Gaudi. And for those of you who had a chance to watch the keynote yesterday when Pat did that live interchange with Matt Hicks, he talked about the excitement that was generated, particularly around Gaudi 3. We actually announced this just about a month ago at Intel Vision in actually my hometown in Phoenix. And we talked about a lot of different things. We talked about the performance that it's bringing, really competitive and compelling performance. It is a third generation architecture, so by comparison to the prior generation of Gaudi, it's bringing 4X the computing capability, 2X the networking performance, 1.5X The memory bandwidth. And it is very competitive versus other solutions that are in market today in both raw performance and particularly performance per dollar in terms of being able to deliver cost-effectiveness. But it's also addressing specific needs in enterprise. So we're expanding the OEM ecosystem. All of the four major OEMs across Dell Technologies, HP, Supermicro, Lenovo, all announce that they would be in market with solutions with Gaudi 3 when we bring it to market later this year in the second half. And we're also expanding the product line as well to support a PCI Express form factor, which is really interesting to enterprises actually, because you can deploy the technology and a form factor that fits into standard servers. And so we know that that's critical, but we also know that there are opportunities to combine the Gaudi accelerator technology with Intel Xeon technology as well. And so when you look at, for example, a really critical workflow in enterprises in the area of RAGs, retrieval augmented generation, harnessing the power of LLMs, but applying these other sources of knowledge, we've demoed ... and actually over in our show floor right around the corner, we've demoed a capability of bringing together both Xeon and Gaudi to service different portions of that workflow. So we really look at it as the future is definitely accelerated compute, but we're bringing multiple different assets to the table.

Paul Gillin

>> What's the difference between an accelerator and a parallel GPU architecture?

Jeni Barovian Panhorst

>> So Gaudi is an accelerator. We also have GPUs in our portfolio as well, and we definitely have a forward-looking roadmap that focuses on GPUs as well. We are offering developers access to the accelerators that we have with open programming models, leveraging open frameworks like PyTorch and ensuring that they can bring their models that they're developing on GPUs today onto those accelerators.

Steven Huels

>> I think it adds a lot of value. So from a Red Hat perspective, we get demands from customers to run across a lot of different hardware footprints. And when you start talking about being able to run these generative AI models maybe in edge locations, not everything's going to require a GPU. There's a lot of technology that's driving these generative AI models to be smaller, more efficient, more effective. That opens up a lot of opportunities for smaller processors, lower power consumption, lower heat footprint, more efficient, and that's, I think, what enterprises are really looking for. But all of that across a consistent stack gives them optionality in how they build and then where they deploy.

>> Yeah, I think in fact, I was going to hit on that because it was funny. We were talking about how things get deployed out to the edge in particular when you're talking about the fact that you need ... latency is a real thing. And especially when you're doing certain data apps and having gen AI being part of a data app and maybe it's actually deployed out to a 5G cell tower, and you want to minimize that. You want to do the inference out at the edge and be able to be responsive as part of an app. Because I think longer term ... And in fact we had Mo on from Red Hat who's engineering head for InstructLab, we had her on prior to that at KubeCon just a few months ago and we were talking about UX and some of the different ways that people are going to interact with these gen AI products, and it's not always going to be through a prompt. But where are you seeing, or how are you seeing and suggesting to enterprises that they really take a strategic approach to gen AI, and what are some of the things that you're seeing and helping them with?

Steven Huels

>> So from a Red Hat perspective, the rate of change in innovation and AI is explosive right now. So companies know they're going to have to take some speculative bets on technologies that might not take off, but they don't want to compromise their entire AI investment. So what we talk to them a lot about is our hybrid cloud story, especially with OpenShift AI, where we can provide a core platform that allows you to build, train, tune, deploy, monitor, and manage those models, but still gives composability. So if you want flexibility to bring in some of those speculative bets, if it doesn't work out, you're not compromising your entire investment, you can still bring in another technology and augment that with even your own innovation and open source innovation. And then underpinning that is a lot of our ecosystem. So we give them the optionality based on the type of workload, the specific use case. We can work with different types of accelerators, we can work on CPU. As much as we talk about generative AI, the bulk of what's out there is still predictive AI, and we can also run those workloads together, so we can do predictive and generative. And so really, that's our core message to them is if you're betting on Red Hat, you're going to be able to have a stable platform for the future that can absorb the changes that are coming and still work with all the innovation that's coming out of the ecosystem.

Paul Gillin

>> Jeni, I'm curious, your title is Data Center AI Product Management. Does that imply that Intel believes that the AI training is going to move largely to the data centers, or is it going to be a largely cloud phenomenon?

Jeni Barovian Panhorst

>> Well, certainly there's training that's happening in cloud data centers, but we're seeing definitely enterprise interest to not just not only harness the capabilities of training in their cloud deployments with the hyperscalers, their hybrid cloud deployments, as Steven was saying, but also we're seeing an increased interest in on-prem deployments for both training and for inference, although inference certainly is dominant in on-prem deployments and certainly at the edge.

>> Yeah, I think again, as we look at it, there's also the whole idea that Intel really does lean into the open aspect of this as well. Help people understand some of the announcements. I think Pat made some announcements recently with the Linux Foundation. What is it that really, Intel is leaning in with open source and with developers?

Jeni Barovian Panhorst

>> Yeah, open source has been critical to certainly everything we've done in partnership with Red Hat over the course of the past 30-ish years or so, and certainly what we're driving now with our approach to AI. I mentioned before taking advantage of open frameworks like PyTorch, model communities like Hugging Face. You mentioned InstructLab. I think there's a lot of really exciting innovation that's going on there in terms of building a community approach to adapting the power of large language models to different enterprise domains. But I think we're alluding to the announcement that Pat made around OPEA, or the Open Platform for Enterprise AI, and we initially talked about that a bit at Intel Vision about a month ago. The community launched in Linux Foundation immediately thereafter. There's about 15 companies that are signed up at this point, including Red Hat. We're really excited to get this community off the ground. And again, it's really trying to take an open and community approach to accelerating that pace of innovation and pace of development so enterprises can really harness the power of gen AI. And so I mentioned before a little bit about RAGs. We're contributing our RAG implementation to OPEAI as one area to focus on kickstarting that. Things are moving very quickly. We're going to be hosting community days next week. So there's an opportunity for anyone to get in on the ground floor and participate as we continue to build out the technical framework and start to build up the working groups there. So definitely a good opportunity for engagement.

Steven Huels

>> And a lot of the work we're doing focuses a lot on the consumability of the integrations. At the end of the day, the thing that's going to make AI successful is customer's ability to consume it. While it's great to create it, and it's exciting to play with a lot of the models, really it's the ability to roll it into production that is going to generate the value. When we talk about the integration we've done with enterprise AI, it really is a full stack integration, where from a customer perspective, it's invisible to you. The components are just there, you're getting access to the optimizations, whether it's at the accelerator layer, at the runtime layer for the model, at the developer layer for the various libraries. So it's an integrated experience that helps you generate value and put those models into production.

Paul Gillin

>> You mentioned earlier that about 40% of PCs shipped this year are expected to be AI PCs. What exactly does that mean? And of course, the desktop is Intel's wheelhouse. How do you look at the future? How do PCs evolve as AI machines?

Jeni Barovian Panhorst

>> I think there's a couple of different ways to look at it. I think there's certainly applications that we're all using every day, if you think about, for example, video conferencing applications and the idea to even just make visual processing much smarter. How do we make things like background images and blur and other enhancements to the ways that we communicate on a daily basis? That's harnessing the capabilities of GPUs and enhanced AI capabilities on an ongoing basis. There certainly is an opportunity for smaller language models running locally on a PC for very low latency and high responsiveness just as part of our normal usage of assistants and copilots on an ongoing basis. And so a tremendous amount of innovation happening there just as part of productivity applications, developer applications. And then that's actually where I wanted to take it next, which is how do we enable developers to move faster? So actually harnessing the GPU capabilities that exist on the platform as a proxy for what they're developing for ultimately in an enterprise data center or in the cloud. So it's a way to get started on your development and build code that ultimately scales to a broader deployment.

Steven Huels

>> And we're seeing the same trend. You mentioned InstructLab, and I'm sure Matt mentioned RHEL AI and what we're delivering there. We are seeing a demand for developers to be able to work with these language models on their PCs and then be able to move those up into production. And so we're taking advantage of all the innovation that's happening on AI PCs. And then when you're ready to move RHEL AI into production, we can roll that into an OpenShift AI and roll it across the enterprise. So we're really starting to mirror what you saw for application development now for AI, and again, taking big steps toward overall consumability.

>> Yeah, no, I think that totally makes sense, and I think that's a great place to leave it, beyond the fact that how do people find out more? I hear there's a blog that they can sign up for, something like that.

Jeni Barovian Panhorst

>> Oh, there's all kinds of stuff. We've got an AI blog that you can sign up for. The other thing that I would encourage developers to check out in particular, we've talked a little bit about Gaudi, we've gotten a lot of questions asked about, "How do we get access to it?" One thing that Pat talked about yesterday as well that I'll emphasize again, is that we do have Gaudi 2, and will have Gaudi 3 instances on Intel Tiber Developer Cloud, empowered by Red Hat and Dell Technologies. And so that is another way for developers to get access and develop for now and into the future. Certainly there's an opportunity to also engage in various communities. OPEAI is one place. But like you said, this whole conversation is about open source and communities, and so lots of opportunities to develop the open approach to enterprise AI in the future.

Steven Huels

>> Yeah. And here, you don't have to go far. I think I saw four or five of the InstructLab sweatshirts wander by. So join the InstructLab community, get started with large language models on your laptop, and then have a look at us at OpenShift AI as well.

>> Not to mention Podman.

Steven Huels

>> Yeah, Podman AI.

>> I guess I'll help you out with that. But hey, thank you both for coming on board. Really appreciate it. You did wonderful, and hope to have you back soon.

Jeni Barovian Panhorst

>> All right. Thank you so much.

Steven Huels

>> Thanks for having me.

>> And thank you for watching this episode of theCUBE from Red Hat Summit and AnsibleFest here in sunny, frigid-ish Denver today. High altitude's getting to me, getting a little wacky. But we're still here. Got a lot more to come. So stay tuned to theCUBE, the leader in high-tech information, analysis and news.