NVIDIA Corporation (NVDA) Earnings Call Transcript & Summary

Thank you for joining this webinar on Programming the Modern Data Center. It's nice to see so many people on, so many of our Inception members interested in hearing about this topic from our experts. My name is Daman Oberoi. I work as a partner manager for the Inception program here at NVIDIA. I'm on the U.S. and Canada team, and I'm part of a much larger global team that supports tech start-ups in all industries. I'll tell you much more about the program here in just a second, but first, I want to cover some housekeeping items. We will be answering questions live, so please submit them throughout the presentation. Check out the related content section for links to slides, blogs and other content that will expound upon what we will discuss in the presentation. If you have any technical difficulties, it can help to close other applications or tabs or refresh this page and re-enter the webinar. Those are common ways to solve technical issues that people sometimes face. As for the agenda today, I'm going to cover an overview of the Inception program. Many of you on this call today may already be in the program. I'd like to cover our latest benefits, which you may or may not be familiar with. Then I'm going to toss it over to Shai and Alvin, who will cover the BlueField DPU and DOCA respectively. We have a special guest, Dave, from Cirrascale, is going to talk about their DPU offering. And then we're going to do that live Q&A at the end. NVIDIA CEO, Jen-Hsun Huang is quoted to have said, "Great innovators recognize the art of the possible." This is a perfect encapsulation of what the Inception program and really all of NVIDIA is about. Our purpose is to create the technology, the tools, the SDKs, the hardware that allow our partners and our customers to take what we've built and run with it. The Inception program has exploded in popularity, as you can see on this slide. We have over 13,000 start-ups now with a cumulative funding over $94 billion, 100 or more countries represented. And the program continues to grow at a very fast pace. The Inception program is designed to help start-ups at any stage. We have companies that join Inception that literally incorporated a week ago. They are just a couple of people who are trying to get started building their technology and need access to training and guidance on where to go. We have companies on the other end of that scale where they are really at massive scale, and they're trying to grow their product in the market. Maybe they've got Series C or D funding or more even. We can help them with market presence and accelerating their solution on NVIDIA technology. And then, of course, the majority of companies fit somewhere in between those 2 extremes, we've got solutions for you as well. We want to help you build and get to market faster. So we give our start-ups a slew of benefits. Starting on the top left corner here, we have the Deep Learning Institute or DLI, which is NVIDIA's platform for training about our SDKs. You can take self-pace training through there or you can sign up as an Inception member with a 50% discount on instructor-led workshops. Moving to the right, we have our preferred pricing on NVIDIA's enterprise-grade hardware. You can log in to your Inception portal, you can find a list of hardware that's available for a discount and how much that discount would be, and you can submit your request right there. We work with AWS as our cloud partner to offer you up to $100,000 in AWS activate cloud credits. And we have a massive collection of SDKs and developer tools, over 150 of them where you can get support at a level -- as an Inception member that other folks using these tools cannot get. You may have heard that NVIDIA now has an offering in the cloud called DGX Cloud. It was just announced at the most recent GTC a few weeks ago. And what that includes is a license to the NVIDIA AI enterprise software suite. It includes access to a base command platform, which is a single pane of glass for you to manage everything on your DGX Cloud instances. It comes with 8 GPUs per cloud instance, 10 terabytes of storage per instance. And then you get a whole support team of account managers, success managers and other experts to support you as you use this service. The list price is just shy of $37,000 per instance per month for all of this offering. But as an Inception member, you also get a 30% discount on that price. And furthermore, if you commit to a longer-term agreement, the price per month comes down even further. NVIDIA LaunchPad is a free program that provides immediate short-term access to the necessary hardware, software and data to experience end-to-end solution workflows in the areas of AI, data science, 3D collaboration and more. You can get access to hands-on labs to test out our various SDKs, might that be RAPIDS or Morpheus or Riva or any number of other SDKs, which you can check out the labs at nvidia.com/launchpad. As an Inception member, you get a special queue to get on to launchpad. And so I recommend that you contact your Inception partner manager if you're interested in a trial on NVIDIA Launchpad. A big part of the benefit of the Inception program is we help you with brand awareness. I'm going to talk over the next several slides of all the ways in which we can help you with that. The first I want to point out is our Accelerated Apps Catalog. This is the website of applications built on GPU by our partners. So each one gets a tile here on this web page where it's got the name of the product, the solution, the offering from our partner. It's got a short description of what that solution is, what it does and then some key features there of NVIDIA technology that enabled particular behaviors within that application and then perhaps, most importantly, a link at the bottom there to your website to go sign up or download or however the customer can access store offering. This slide covers the range of items that we typically help our partners with when it comes to brand awareness and market exposure. So on the far left here, we've got the NVIDIA blog. There's actually 2 blogs that we support. There's a corporate blog, which we, with your input, of course, write the article. And it really is targeting a C-suite audience. These are business development types of hubs. And then, we have a technical developer blog that is really, obviously, a more technical angle on a particular topic. So it might include code walk-throughs, code snippets and other such information that is really, really pertinent for our developer audience. We also have an AI podcast. If you haven't checked it out, strongly recommend you go check it out. We have Inception partners featured on there from time-to-time, and you could be one of those companies that is included on the NVIDIA AI podcast. We have a number of marketing assets that we create. They all have the sort of the same layout and same approach and how we go about doing them, but they're all written with a slightly different angle. And what I'm referring to here is a solution brief or what we call a solution showcase. There is also a customer success story. And then there's a use-case study. And they're all different angles on essentially the same type of content, which is that on the majority of the page, it'll talk about your solution and the problems that it addresses. And in the side bar, it will talk about the NVIDIA technology that helped enable that solution. With a customer success story, very similar, but now it becomes a 3-party story where your customers' logos on the page, your logos on the page, our logos on the page, we talk about the customers' problem that your solution helped to address, and again, in the sidebar, NVIDIA technology that was used to address it. And then the last tile here talks about industry publications, which come through from time-to-time. There's an opportunity there. But the more likely and more common situation is that the marketing team comes to us quite regularly and says, hey, we're doing a spotlight. And so we'd like to know which of your partners would be a good fit for this spotlight. So for instance, there was a speech AI spotlight a month or 2 ago. And the marketing team came to us and said, hey, we're looking for partners that are doing speech AI using NVIDIA technology. So all of us partner managers, we connect the marketing team to the key partners that are relevant for that particular story, that particular use case or what have you. So these are the various ways in which we can help you with your brand awareness. But it doesn't end there. We also do a lot of support for events. Specifically, we can provide you assets that show that you're part of the NVIDIA Inception program. So this little green tabletop sign, it's very easy to e-mail this over to you. And it really just gives you that extra stamp of credibility that you are partnered with NVIDIA and that visitors to your booth should really take note of that. Amplification on social media is another thing that we do for our partners very commonly. The easy way to go about this is any post that you have that has to do with NVIDIA technology or how NVIDIA helped you in some way to enable something, you can include the #NVIDIAInception on that post or you can @NVIDIAAI or another great way is just to get in touch with your inception partner manager who can help navigate this process with you. The 3 platforms that we primarily use for NVIDIA social media are Facebook, Instagram and LinkedIn. I'd be remiss if I didn't mention GTC, which is the GPU Technology Conference that NVIDIA holds, typically twice per year. We have -- at the last event, just last month, we had over 250,000 registrants. And so there's a lot of visibility if your solution, if your company is highlighted in the keynote or elsewhere throughout the conference. So this is a slide with 10 of the companies that were highlighted in a 2022 keynote. And that keynote was viewed by millions of people. So again, it's incredible visibility should your solution be featured in GTC. So with that, we're going to transition now. I'm going to hand it over to Shai, who's going to get into the meat of the technical presentation.

Hi, and welcome to this webinar, Programming the Modern Data Center. My name is Shai Tsur. I'm the networking lead for NVIDIA's Inception program, helping start-ups and inception members adopt NVIDIA's BlueField and DOCA technology. With me is Alvin Clark, senior solution architect on the Inception team. So in this webinar, we're going to talk about a revolution that's happening with modern applications and how this revolution is really driving fundamental changes in the architecture of cloud data centers. So we'll have a look at what is happening in the technology landscape and how these changes are really driving the way data centers are built and operated. Then we'll talk about the data processing unit with DPU and see how that fits into these new changes. We'll go into details of NVIDIA's BlueField-3 DPU before talking about the DOCA stack, which is the software stack that is used to program the DPUs. Then we'll talk about use cases, what people are doing with BlueField and DOCA and talk about the service that we have to offer DPU access in the cloud. So before we talk about the DPU and DOCA, let's take a minute to look at some of the changes in the technology landscape. So we've talked about a new wave of applications that we've seen in the last couple of years, and these are really transforming the industry as a whole. Most people should be aware of the generative AI-type applications. ChatGPT, in particular, has been all over the news as the fastest-growing application in the history of the Internet. And the idea behind the generative AI is you take a large language model. And based on text prompts from the users, create new media or code snippets or video or graphics based on that. It's a truly transformative and revolutionary technology. Next is data science or what used to be called Big Data, which has been around for a few years, and has really changed the way that organizations do their decision-making. It helps them drive data-driven decisions for things like performance optimization or gaining market traction. And then there's the metaverse or what we call the 3D Internet, which is revolutionizing the worlds of manufacturing with Digital Twins and is set to revolutionize both the social Internet and gaming by creating 3D worlds. And of course, NVIDIA has a -- the Omniverse platform for managing metaverse applications. So what all these types of technologies have in common besides being incredibly disruptive is that they all-demand huge amounts of computing to run at scale. And this presents a host of challenges not only for the cloud service platforms that have to run these technologies, but also the application developers who are looking to take advantage both of these technologies and the new way the data centers are being architected. So if we want to understand this better, let's take a look at how data centers have actually evolved over the past couple of decades. So in the past, data centers were generally on-premises, right? This is before the cloud revolution. They consisted of racks of either general purpose servicers or monolithic devices that were used for running the infrastructure tasks of the organization. I'm talking about things like firewalls and routers and storage devices. Now this worked until you needed to really start scaling out. This set up didn't scale well, and that's really where we got the cloud revolution from. Now the big change in the cloud revolution was going from monolithic-type applications to what we call the software-defined data center. Now these tasks of networking, storage, compute and other infrastructure tasks, instead of running on single-purpose devices, were turning to software and run on the same general purpose service servers that we're running the business applications. This provided a lot of flexibility because now, if you had more demand for, say, storage, you didn't have to put in a whole new storage device or a whole new router to meet temporary demands, you can just add more software. The problem with this approach is that it creates a lot of overhead on these servers. In fact, managing a modern data center just from the infrastructure side takes up 30% of the CPU power of these servers. And we can see that in this dark gray box. Now this problem has continued to compound as we moved on to where we are today, which is a situation of disaggregated micro services, containers where everything is scaling out, but applications, instead of running on one server, may be running on different servers or different racks of servers. And this creates a whole new range of overhead that you need to manage. And all these changes are being accelerated by adding GPU into the mix. So this creates even more performance needs. And really, this is where the DPU comes in to answer some challenges. So what are these problems that we've been talking about that create challenges for modern data centers? Well, first, the modern applications are operating at a scale that we've never seen before. These are huge containerized and distributed across the data center, and they're also decoupled from the underlying infrastructure. As a result, you get massive increase in networking between the servers inside the data centers. That's what we call east-west traffic. Now at the same time that you have this additional overhead, you have applications such as AI, IoT, 5G wireless connectivity and others that have super stringent performance and latency requirements, and these are pushing the compute and the networking technology to its real limits. Now at the same time as the performance requirements are increasing, the cost to power data centers is also increasing. So we've seen that there's been a real spike in energy prices over recent years. And not just energy prices, there's a realization in the data centers, by consuming so much energy, are having a real environmental impact. So there's a need to really power data centers as efficiently as possible. Now remember, the software-defined data center, like we saw, utilizes something like 30% of its processing power just to handle the infrastructure. And so when you're trying to maximize the performance of these servers in order to kind of optimize your power spending, this becomes a real drag because you're consuming CPU cores that could be running business applications or mission-critical applications and instead using them to run infrastructure. The same principle applies to other applications, which are things on the far edge or in areas like autonomous vehicles or places that are limited either by their physical size or by their resources. We really need to maximize the performance and get the most performance in -- out of whatever hardware you have in that space. And last but not least, there are huge security implications for all this. So as the business applications and the management are running in the same servers, what it means is that the security policies is sharing a trust domain with a host of tenant applications. And any vulnerability in one of those tenant applications running on the servers can infect the security management of the whole data center, so a security breach in one application could spread throughout the data center. So you really need to be able to isolate the security management from the rest of the host. So these are the challenges that the DPU looks to solve. Another way of looking at it is that these modern applications are becoming so large and so intensive that they simply don't fit into a single server or, increasingly, even racks of servers. So instead -- instead of the server being the unit of compute like it used to be, we're now looking at a new world where the data center itself really is the unit of computing. So in order to make this model work, you need what we call the 3 pillars of the data center, the CPU, the GPU and now the DPU. So the CPUs are the workhorse of the data center. They are -- they excel at serial processing, which you need for running the computer operating systems, the power of the data centers and really enable the multitasking at the compute and data center levels that you need for cloud computing. The GPU, of course, excels a parallel processing, which is optimizing -- optimized for accelerating workloads such as graphics and AI, which are really crucial to this new class of application. And now we have the DPU, which is the programmable processor that specializes in offloading and moving the data from these modern data center workloads and tasks. And this is becoming increasingly important, not just for generative AI and Omniverse, but really everything that we're seeing in the modern world such as streaming media, 5G, cloud, IoT and you get the idea. So now that we've gotten the landscape, let's talk about -- in detail about NVIDIA's DPUs, which we call BlueFields. So we've been very pleased to announce recently the release of the third generation of BlueField DPUs, the BlueField-3, which is really a complete infrastructure and compute platform. So let's have a look at what's in a BlueField-3 DPU card. So the BlueField has a number of different components. The first off is a CPU subsystem. Now this includes 16 Arm A78 cores that can be used to run the management of the application task that we talked about -- the infrastructure task that we talk about, the networking, security and storage, directly on the DPU, instead of having to run them on the host but these arm cores are not just there to run infrastructure tasks. They are general-purpose CPUs and can be used for running any other workloads that have been ported to the arm architecture. And this provides a real flexibility about what you're running in the host and what you could run outside the host, which you need for that scale and performance. The next component is an NVIDIA ConnectX-7 SmartNIC, which provides up to 400 gigabit per second of throughput. This is responsible for handling the networking tasks we talked about efficiently and executing the policies that have been defined by the software-defined networking applications. So this is the main networking piece that you need for the data center. But -- so these 2 components, the CPU cores and the ConnectX-7 have been around on previous generations of BlueField. What's new is the DataPath Accelerator, which is called the DPA. This is a highly parallel and very performant I/O processor, and it's designed to greatly accelerate things like device emulation, network flow processing, intensive I/O applications and many more. So again, a purpose-built accelerator for specific networking tasks that really accelerate those tasks. And in addition to these are additional acceleration engines that have been -- for workloads that have been offloaded to the DPU. These acceleration engines greatly improve performance for a whole range of applications, things like malware detection, video streaming and many others. So if you take all these things together, you can really think of the BlueField as kind of a specialized server that sits in front of the general server and is there to handle the data center infrastructure tasks in a superior manner. So BlueField-3 is currently sampling with very select clients but will be generally available starting the second half of this year. So now let's look at what the DPU actually does. When we talk about the BlueField and its advantages for developers, we really talk about its ability to do 3 things: to offload, to accelerate and to isolate workloads. So if we go back to this idea of the evolution of the data center, we saw that during the rise of the cloud, we went from having monolithic devices to handle application management to having the same infrastructure tasks being done in software running on the host server alongside the other business tasks, which generally run in virtual machines or containers. So in this case, everything in blue is our host server. Now in order to scale and increase performance, we're looking to -- instead of running these infrastructure tasks on the server, run them on more dedicated hardware devices. In the past, we were able to do a lot of the networking pieces on NVIDIA's ConnectX-7 SmartNICs, which were the leaders in the field and continue to be the leaders in the field, but these are made for executing networking tasks. So you define the networking policy in the host with the software-defined networking. And those policies are executed by the SmartNIC, which greatly improves networking performance. But with the DPU, you don't just offload the networking. You can take all these infrastructure tasks, which have been defined in software. And instead of running them on the host, run them on the DPU itself here in purple. And then the cores that would otherwise be used for running those tasks in the host can now be used for running additional virtual machines or containers, and this really improves the utilization of the host resources. So that's the offload piece. Now the acceleration piece we started talking about in the previous slide whereas the hardware that's available on the DPU, whether it's the -- not just the ConnectX-7, but the DPA and the other specialized accelerators are really here to make all these tasks perform as quickly as possible. And this is the acceleration piece, which, again, is crucial for the performance requirements of modern applications. And the third piece is isolation. So the DPU has -- takes advantage of the Arm Zero Trust security policy, which enables you to create a security gap between the DPU itself and the host. So as we mentioned before, one of the big threats for a modern data center is that a security breach that comes from one of these tenant tasks would, in fact, be the software-defined security management and spread out through the data center. By offloading that piece to the DPU, the software-defined security is immune from what happens up in the host. And therefore, it provides a whole level of software management. So what I'd like to take -- what I'd like you to take away from all this is the DPU really provides its value by doing these 3 pieces to kind of offload CPUs, accelerate and isolate. So if we can summarize, by creating this architecture that does offload, accelerate and isolate, the DPU is the answer to those problems that we talked about before. So if we're looking at the need for accelerated performance by creating a whole range of built-in accelerators and also acceleration in the DOCA SDK, which we'll talk about in a minute, the DPU allows applications to run their infrastructure in a greatly accelerated manner. I should also mention that applications that are built to run on the DPU can also take advantage of these acceleration performances. So it's not just the infrastructure management, it's the actual application itself that can be accelerated using the DPU. The idea that you're offloading pieces to the DPU allows you to free up cores that can run additional VMs or containers and other applications, right? So this allows you to maximize the performance power of the CPU in the host server, which we said is crucial for optimizing the power management of the data center. And the Zero Trust security that you get from moving the -- all the software-defined security tasks to the DPU, which is air gapped from the host server, provides a whole new level of comprehensive data center security while retaining the flexibility that you have from software-defined security. And of course, one thing that's also again worth remembering is that the CPU cores on the DPU, the 16 Arm A78 cores, are really kind of a general-purpose server in and out [ of themselves. ] And they allow application developers to write applications that perform consistently on the DPU. So again, the data centers become -- now with the data center is the unit of processing, the DPU will really play in a crucial role in providing this. So now that we've talked about the DPU and its role in the modern data center, I'm going to turn it over to Alvin to talk about the DOCA SDK.

Thanks, Shai. BlueField DPUs and DOCA enable developers to transform their data centers into state-of-the-art virtual private clouds with public cloud scalability that is accelerated, fully programmable and secure. In our vision, DOCA unlocks data center innovation by enabling developers to rapidly create applications and services on top of DPUs, similar to CUDA for GPUs, leveraging industry-standard APIs. In support of the BlueField-3 production availability, NVIDIA has announced DOCA 2.0. DOCA is the open cloud SDK and acceleration framework with NVIDIA BlueField DPUs. Over 4,700 DOCA developers are using DOCA to unlock data center innovation by enabling the rapid creation of applications and services that will run on top of BlueField DPUs. With DOCA's extensive libraries, drivers and APIs, DOCA is a one-stop shop for BlueField developers, and it's the key to accelerating infrastructure services in the cloud. DOCA offers forward and backward compatibility across BlueField generations with more than 500 APIs to simplify the developer journey. DOCA targets 2 types of personas, the application developers and the IT administrators, which can also include DevOp engineers. From the point of view of the developers, the DOCA SDK includes all the components that are required to create and build application. This includes sources, development tools, documentation and tools like the SDK manager that can easily deploy the development environment and can also flash and install the software on your local DPU. The DOCA SDK can be consumed using a development container with Arm emulation support that can run on an x86 machine or directly developed on the Arm course of the DPU itself. Also available via the DOCA documentation are examples: API compatibility information and best practices outlining the most efficient way to leverage the local libraries and drivers. On the flip side, the IT administrative persona is different. They're not necessarily expected to create applications, but to leverage existing applications and services coming from NVIDIA and third parties. In this case, what they need is the DOCA run time, which includes all the binaries for libraries and drivers needed to execute the accelerated applications created by the developers. In addition, DOCA run time brings DOCA services such as DOCA telemetry that can give real-time information on the performance of the DPUs to administrators about DOCA services. DOCA services greatly simplifies the deployment of DPU-accelerated applications across the data center. Utilizing containers, users can pull third-party applications, applications from the NGC catalogs or from private repositories and deploy them across various environments and multi-generations of the DPU. It's one of the great ways that the DOCA run time can help users scale their applications across the data center or multiple data centers. Now I'll hand it back to Shai. Thank you.

Okay. So we've talked about what the DPU and DOCA is. Now let's talk about use cases because, really, this is great technology. But what can we do with it? Now as we mentioned, there are more than 4,700 developers developing BlueField applications today. So we can't get into every single use case. But let's talk about some of the general areas and get some examples of people doing things in those areas. Let's start with cloud computing, because as we've talked about throughout this presentation, the DPU is here to revolutionize the way that the cloud data centers are being architected. And so it makes sense that we're working very closely with the companies that are helping power these data centers, really operate these data centers. We're talking about the cloud OS partners such as VMware and Red Hat. VMware announced its Project Monterey, which allows organizations to run ESXi, which is the leading operating system for running virtual machines and is now also running containers directly on the BlueField, again, for the scale and flexibility that we need for modern data centers. With Red Hat, their OpenShift platform, which is a leader in container management, is also being run on the BlueField. And we're working with many other partners such as Canonical and SUSE, albeit the companies really working on providing the basis for the cloud. Cybersecurity is a huge area, and we're seeing many, many cybersecurity companies taking advantage of all the different features both in BlueField and of DOCA. One that we can certainly mention is Palo Alto Networks, which is a leader in this field. For their next-generation firewall, Palo Alto has developed a technology called Intelligent Traffic Offload, which takes advantage of the BlueField. With the ITO solution, data flows into the BlueField and the BlueField inspects the packets coming through and makes the decision is this traffic that is safe or traffic that needs to be looked at by the firewall. Save traffic, Netflix video streams, for instance, get routed directly through to their destination without ever having to see the host at all. Anything that is suspicious gets routed up to the firewall, which is sitting in the host. By adopting this approach, Palo Alto Networks has been able to increase its data throughput by more than 5x. So again, a real performance boost. For Telco and Edge applications, we're actually seeing a similar revolution that is what we're seeing in the cloud data centers where the Telco companies are increasingly rolling out infrastructure based on general purpose servers to greatly expanded areas rather than concentrating them in their data center core. And for this, as we've seen, the BlueField excels. I can point out some work that we've been doing with Mavenir called for a hyper-converged 5G AI platform that takes advantage not only of the hardware offload capabilities in the BlueField, but also a DOCA library called Firefly, which is responsible for maintaining ultra-reliable timing and synchronization between servers, which is something that you really need for Telco applications. On the storage front, VAST Data is using NVMe-over-fabrics offload, which is helping drive their new series platform. Series provides a hyperdense storage that's both highly performant and power efficient. So again, taking advantage of BlueField for power optimization and scale. We'll see an additional use case in a second using NVMe-over-fabrics by Bloombase. And then for media streaming and CDN-based applications, I want to call out an additional library in DOCA called Rivermax, which leverages the power that's both in ConnectX and BlueField for hardware streaming, which enables direct data transfer to and from the GPU. This helps deliver best-in-class throughput and latency with minimal CPU utilization for streaming workloads. And this can be looked at not just for media streaming, we can think about this for uses in gaming. And increasingly, we're starting to look at it also for its [ abilities ] in generative AI video. So as I said, there are many, many use cases here. But as a general rule, what I'd like you to take away from this is when you're thinking about is BlueField right for me and how can I use BlueField, I'd like you to think about the different pieces of the applications you're building and where these components live and where they're processed. So how does [ the data get ] in and out of your software? And how much CPU is spent to support those functions? Are the services running in a centralized place and be a bottleneck that could be distributed to the DPUs? Think about these thoughts, and then I believe that you'll find that there's a space for the BlueField to improve that performance. And now let's take a look at some specific use cases of Inception partners working with BlueField and with DOCA. So let's start with an Inception partner called L7 Defense out of Israel who were providing an API security offering, which leverages the offload capabilities of the BlueField. So L7 has a solution called Ammune, which is here to answer the problem of API security. Now as you may know, APIs are everywhere in the modern Internet. Any transaction that you do for e-commerce or banking or really anything that's done over the Internet these days involves tens, if not hundreds, of API calls to different servers. So as a result, API abuse is becoming an emerging attack vector for cyberattacks. So L7's Ammune is an autonomous AI platform that handles the mapping, detecting and blocking of API cyberattacks. So they have a 2-part solution. One component is an AI/ML piece that runs in the host, which aggregates and analyzes traffic coming through, constantly looking at this traffic and creating profiles of what malicious APIs look like. So based on those models that it's constantly refining, it develops security policies to protect against those attacks. The second component is a real-time AI agent that actually performs the traffic filtering. So it looks at the traffic profiles that the first component creates, investigates the traffic, lets the safe traffic through and blocks malicious traffic. Now without the DPU, both pieces, the analytics and the filtering piece of malicious APIs need to be done on the host, but L7 has offloaded the real-time AI to be able to run on the BlueField. And now the BlueField cards rather than the host become the security edge. By doing this setup, by deploying the solution on the BlueField, the L7 shows that the CPU utilization for their application decreases by 95%. So when we're looking at reclaiming overhead from the host, this is incredibly significant. But it's not just about improving the performance. By offloading the solution to the DPU, they can show that their API defense can now be used not just in cloud data centers, but also in things like autonomous cars and remote locations. And you're really kind of expanding the protective edge out all over the place. What I like about this is it's a great example of how a Inception partner can use BlueField to really expand the types of offerings that they're offering and penetrate new markets. Okay. Our second example of an Inception company is Bloombase. And Bloombase is using the BlueField DPU to answer a different type of security problem, and this is what they call data addressed. So one of the emerging challenges for security companies is the growth of the quantum computing. Quantum computing allows malicious actors to target existing firewalls with much more effective brute force attacks, which means that if you're looking forward in time, you really can't rely on the firewall to be your last line of defense. And you need to look at the situation of, okay, if the malicious actors manage to penetrate the firewall into the organization, do they have access to secure information contained on files in the storage? So what Bloombase has developed is StoreSafe, which is what they call a storage firewall or storage proxy that in the data center sits between the application servers and the storage servers in your data center. And it's responsible for looking at the data that flows between one to the other and making sure that the sensitive information, what they call the crown jewels of your information, are encrypted. That way, even if there is a break in the security and malicious actors get access to the storage, the sensitive information is encrypted. So how Bloombase StoreSafe works is when an application writes data to the storage firewall, if it's sensitive data, the Bloombase will encrypt it on its way to being persisted in the storage servers. And then when the application needs to request the data again, it decrypts it back into plain text before sending it to the application server. So this is kind of a very effective way of making sure that those sensitive files, once they get into storage and are persisted in storage, are safe. Bloombase is taking advantage of those NVMe offloads that we talked about with VAST earlier to greatly accelerate and offload their -- this capability. And by using the BlueField, they managed to reduce their I/O latency by 75% while increasing their throughput by 5x. These are very, very clear performance gains. And in addition, by offloading it to the BlueField, they're reducing the usage on the CPU by 94%. So again, this is offloading and accelerating. In addition, Bloombase is using a second NVIDIA technology called Morpheus. And Morpheus is a pretrained AI model for different things. In this case, using it for sensitive information detection. So as we said, one of the goals here is to make sure that your sensitive information is encrypted when it's on the storage. But the challenge is defining what is actual sensitive information, which data files need to be encrypted, which ones don't need to be encrypted. And this is a very laborious process that Bloombase is automating using these pretrained AI models. So Morpheus has models to detect what is sensitive healthcare information or credit card numbers or social security numbers, all sorts of, again, sensitive information. And as the data is being stored in the storage via Bloombase, the Morpheus helps the Bloombase StoreSafe understand that this is sensitive data, please encrypt this on the storage. So by using Morpheus for classification, they've managed to decrease the mean time to save the sensitive data by 99%. So again, here is an example of an Inception company that is using both the offload and acceleration piece to create new value for its customers. And Bloombase is about [ to GA the solution at RSA. ] I'd like to take a moment to talk about benefits that Inception members get -- getting them started on their BlueField journey. We currently have the ability to provide a pricing discount on the first few BlueField or ConnectX cards that Inception members want to try out in their data centers. And this is done by ordering those through one of our distribution partners. We will send the distributor partner a reduced-price quote, which they then pass on to the Inception members. And if you're interested, you can contact your Inception manager to discuss. The other benefit that we wanted to talk about is the Cirrascale Cloud DPU. Now this launched at the end of 2022 and is a way for companies or developers looking to try out the BlueField and DOCA to be able to kind of start working on that without having to buy cards to put into their own labs. So this is great if you want to get your hands -- get some experience working with BlueField to see what it does. It's also great if you are ordering hardware and want to kind of take advantage of the time while the hardware arrives to get started -- to reduce the time that you need to spend developing once you get it set up in your lab. We're going to hear a presentation from Cirrascale in a minute talking about the offering and their special free trial and special pricing for Inception members. [Presentation]

So let's summarize what we've learned today. There are big changes in the technology landscape driven by modern applications, which are really pushing data centers to their physical limits. And at the same time, as the data center architecture changes to meet the scale, power, efficiency, security challenges, there's a real need to offload, accelerate and isolate infrastructure tasks. And really, the one thing I want you to take away from this presentation is offload, accelerate and isolate. This is the value of the DPU and DOCA because that's the way to really create applications that can take advantage of modern data centers. BlueField-3 is the industry's leader in DPUs. We have -- it's turbocharged for offering compute, networking, acceleration, security isolation. It's the world leader in infrastructure task for the data center. The DPU has a wide range of use cases. We've seen cloud, security, storage, gaming, 5G, wireless, really many, many, many more. So it's generally applicable to all sorts of applications. And it calls on DOCA, which has a very developed stack of drivers and libraries that are optimized for leveraging BlueField. And again, by using DOCA, not only can you build efficient DPU optimized applications, but you're also protecting your investment by making sure that there's forward compatibility for the next generations of BlueField. So finally, I'd like to point out some BlueField and DOCA resources that will be available as links with this webinar. The NVIDIA DOCA developer form has a wealth of information, not only documentation, but also code samples and even sample applications to help you get up and running with DOCA. We have a DLI course, a developer course about the -- an introduction to DOCA for DPUs that's available for free. We did a podcast last year on Kernel of truth, talking about demystifying the DPU and DOCA that's really worth a listen. And we've recently put out a white paper talking about the power efficiency and the power savings gains you get using BlueField. We also have a large number of developer blogs both on DOCA and on BlueField.

Thanks to our presenters for that insightful presentation. I want to move now to the live Q&A here. We have Shai and Alvin and Dave all here available to answer the questions. If you haven't entered your question yet, please do so, and we will get to it in the time remaining. We will try to get to as many as we can in the time remaining today. So I want to start off here with the first question. Shai, this is for you. The question is we develop for the cloud. What is the adoption of BlueField for cloud providers?

Thanks, Daman. This is a great question. So we are in the process of really kind of expanding the entry of BlueField into cloud environments. We -- the bigger news that came out of our recent GTC session is that Oracle has announced that it's going to be the major -- or the first major cloud service provider to start using BlueField-3 coming up this year. So that's going to be very exciting. We're talking to other -- the major cloud providers and there will be other ways to access BlueField cards in the cloud. And of course, currently, Cirrascale is a great, great option for that, especially they have their -- the BlueField-2 cards. Again, if you're developing on BlueField-2 with DOCA, everything you could be doing is going forward compatible for BlueField-3. So if you're waiting for -- if you think BlueField-3 might be kind of long-term way to go, but want to get started with BlueField, again, I encourage you to contact Cirrascale directly or contact your Inception manager.

Thanks, Shai. Alvin, the next couple of questions are for you. The first one is, is DOCA free? Are you on mute, Alvin? Okay. Well, if Alvin is unable to speak right now, Shai, are you able to jump in here?

Yes. I mean DOCA is free. We do ask that you register at the NVIDIA developer site, which is in the links -- which will be in the webinar. It is a free download. In fact, all our DOCA resources are free.

Excellent. Alvin, are you back? Okay. Sounds like Alvin must be having technical difficulties. Shai, hopefully, you can answer this question in his place. The question is, what is the flow for developing a DOCA app on x86?

So Alvin's actually got more experience about this, but we do have different ways of doing the development. This can be done either -- one of the most straightforward ways since a lot of people are developing on x86 environments, one straightforward way is to use the x86 Arm emulation environment that comes in the DOCA SDK and then that can be used for porting it into an Arm-based environment. You can, of course, also develop on Arm-based equipment, Arm-based servers, which are available. And using the DOCA SDK, you can also develop directly for the BlueField itself. I don't know if Alvin is back.

Yes, I...

Give more color?

Yes. No, that's great. Yes, absolutely. The emulation development container that allows you to run and develop on the x86 machine through the container is a great way to get started, [ has all the ] dependencies. And then, of course, you can develop directly on the DPU itself as well as any, like, Arm cross compilation tools out there that you might...

Excellent. Well, that about wraps things up for us today. That's all the time we have for questions. I want to thank our presenters again for the valuable information here. Please check out the related resources if you haven't done so already. Thanks, everyone, for attending. We look forward to hearing more from you soon. Thank you.