WEBVTT Kind: captions Language: en 00:13:29.708 --> 00:13:31.844 I am a creator 00:13:39.151 --> 00:13:41.353 Blending art and technology 00:13:48.160 --> 00:13:50.729 To immerse our senses 00:13:57.670 --> 00:13:59.138 I am a healer 00:14:01.540 --> 00:14:03.676 Helping us take the next step 00:14:08.614 --> 00:14:10.382 And see what's possible 00:14:19.792 --> 00:14:21.660 I am a pioneer 00:14:25.364 --> 00:14:27.233 Finding life-saving answers 00:14:33.272 --> 00:14:35.841 And pushing the edge to the outer limits. 00:14:41.513 --> 00:14:43.082 I am a guardian 00:14:44.917 --> 00:14:46.585 Defending our oceans 00:14:52.157 --> 00:14:55.394 Magnificent creatures that call them home 00:15:01.433 --> 00:15:03.235 I am a protector 00:15:06.272 --> 00:15:09.775 Helping the earth breathe easier 00:15:16.849 --> 00:15:19.852 And watching over it for generations to come 00:15:23.589 --> 00:15:25.224 I am a storyteller 00:15:27.693 --> 00:15:29.795 Giving emotion to words 00:15:31.830 --> 00:15:33.365 And bringing them to life. 00:15:35.601 --> 00:15:39.138 I am even the composer of the music. 00:15:49.381 --> 00:15:51.116 I am AI 00:15:51.917 --> 00:15:59.959 Brought to life by NVIDIA, deep learning, and brilliant minds everywhere. 00:16:06.865 --> 00:16:09.735 There are powerful forces shaping the world's industries. 00:16:10.302 --> 00:16:18.143 Accelerated computing that we pioneered has supercharged scientific discovery, while providing the computer industry a path forward. 00:16:18.911 --> 00:16:22.982 Artificial intelligence particularly, has seen incredible advances. 00:16:23.482 --> 00:16:29.355 With NVIDIA GPUs computers learn, and software writes software no human can. 00:16:30.022 --> 00:16:35.894 The AI software is delivered as a service from the cloud, performing automation at the speed-of-light. 00:16:36.729 --> 00:16:44.803 Software is now composed of microservices that scale across the entire data center - treating the data center as a single-unit of computing. 00:16:45.604 --> 00:16:50.999 AI and 5G are the ingredients to kick start the 4th industrial revolution 00:16:50.999 --> 00:16:54.771 where automation and robotics can be deployed to the far edges of the world. 00:16:55.614 --> 00:17:02.421 There is one more miracle we need, the metaverse, a virtual world that is a digital twin of ours. 00:17:03.422 --> 00:17:07.960 Welcome to GTC 2021 - we are going to talk about these dynamics and more. 00:17:09.194 --> 00:17:11.130 Let me give you the architecture of my talk. 00:17:11.764 --> 00:17:17.936 It's organized in four stacks - this is how we work - as a full-stack computing platform company. 00:17:18.604 --> 00:17:26.311 The flow also reflects the waves of AI and how we're expanding the reach of our platform to solve new problems, and to enter new markets. 00:17:27.212 --> 00:17:35.721 First is Omniverse - built from the ground-up on NVIDIA's body of work. It is a platform to create and simulate virtual worlds. 00:17:36.422 --> 00:17:43.495 We'll feature many applications of Omniverse, like design collaboration, simulation, and future robotic factories. 00:17:44.329 --> 00:17:47.800 The second stack is DGX and high-performance data centers. 00:17:48.467 --> 00:17:56.975 I'll feature BlueField, new DGXs, new chips, and the new work we're doing in AI, drug discovery, and quantum computing. 00:17:57.810 --> 00:18:00.512 Here, I'll also talk about Arm and new Arm partnerships. 00:18:01.313 --> 00:18:08.320 The third stack is one of our most important new platforms - NVIDIA EGX with Aerial 5G. 00:18:09.254 --> 00:18:14.159 Now, enterprises and industries can do AI and deploy AI-on-5G. 00:18:15.060 --> 00:18:20.332 We'll talk about NVIDIA AI and Pre-Trained Models, like Jarvis Conversational AI. 00:18:21.133 --> 00:18:27.639 And finally, our work with the auto industry to revolutionize the future of transportation - NVIDIA Drive. 00:18:28.273 --> 00:18:32.878 We'll talk about new chips, new platforms and software, and lots of new customers. 00:18:33.212 --> 00:18:34.146 Let's get started. 00:18:35.380 --> 00:18:39.952 Scientists, researchers, developers, and creators are using NVIDIA to do amazing things. 00:18:40.085 --> 00:18:50.162 Your work gets global reach with the installed base of over a billion CUDA GPUs shipped and 250 ExaFLOPS of GPU computing power in the cloud. 00:18:50.462 --> 00:18:57.202 Two and a half million developers and 7,500 startups are creating thousands of applications for accelerated computing. 00:18:57.503 --> 00:19:03.041 We are thrilled by the growth of the ecosystem we are building together and will continue to put our heart and soul into advancing it. 00:19:03.642 --> 00:19:10.349 Building tools for the Da Vincis of our time is our purpose. And in doing so, we also help create the future. 00:19:11.183 --> 00:19:16.755 Democratizing high-performance computers is one of NVIDIA's greatest contributions to science. 00:19:17.523 --> 00:19:22.082 With just a GeForce, every student can have a supercomputer. 00:19:22.082 --> 00:19:28.593 This is how Alex Krizhevsky, Ilya, and Hinton trained AlexNet that caught the world's attention on deep learning. 00:19:29.368 --> 00:19:34.106 And with GPUs in supercomputers, we gave scientists a time machine. 00:19:35.073 --> 00:19:40.979 A scientist once told me that because of NVIDIA's work, he can do his life's work in his lifetime. 00:19:42.214 --> 00:19:43.715 I can't think of a greater purpose. 00:19:44.516 --> 00:19:46.785 Let me highlight a few achievements from last year. 00:19:47.719 --> 00:19:50.622 NVIDIA is continually optimizing the full stack. 00:19:51.156 --> 00:19:56.094 With the chips you have, your software runs faster every year and even faster if you upgrade. 00:19:56.828 --> 00:20:05.589 On our gold suite of important science codes, we increased performance 13-fold in the last 5 years, and for some, performance doubled every year. 00:20:06.305 --> 00:20:12.344 NAMD molecular dynamics simulator, for example, was re-architected and can now run across multiple GPUs. 00:20:13.111 --> 00:20:22.921 Researchers led by Dr. Rommie Amaro at UC San Diego, used this multi-GPU NAMD, running on Oak Ridge Summit supercomputer's 20,000 NVIDIA GPUs, 00:20:22.921 --> 00:20:27.960 to do the largest atomic simulation ever - 305 million atoms. 00:20:28.160 --> 00:20:33.665 This work was critical to a better understanding of the COVID-19 virus and accelerated the making of the vaccine. 00:20:34.700 --> 00:20:39.238 Dr. Amaro and her collaborators won the Gordon Bell Award for this important work. 00:20:39.972 --> 00:20:47.012 I'm very proud to welcome Dr. Amaro and more than 100,000 of you to this year's GTC - our largest-ever by double. 00:20:47.846 --> 00:20:52.290 We have some of the greatest computer scientists and researchers of our time speaking here. 00:20:52.290 --> 00:21:00.532 3 Turing award winners, 12 Gordon Bell award winners, 9 Kaggle Grand Masters - and even 10 Oscar winners. 00:21:00.993 --> 00:21:04.796 We're also delighted to have the brightest minds from industry sharing their discoveries. 00:21:05.330 --> 00:21:13.171 Leaders from every field - healthcare, auto, finance, retail, energy, internet services, every major enterprise IT company. 00:21:13.939 --> 00:21:20.557 They're bringing you their latest work in COVID research, data science, cybersecurity, new approaches to computer graphics 00:21:20.557 --> 00:21:23.317 and the most recent advances in AI and robotics. 00:21:23.849 --> 00:21:31.923 In total, 1,600 talks about the most important technologies of our time, from the leaders in the field that are shaping our world. 00:21:32.491 --> 00:21:33.492 Welcome to GTC. 00:21:34.459 --> 00:21:37.763 Let's start where NVIDIA started...computer graphics. 00:21:38.397 --> 00:21:46.102 Computer graphics is the driving force of our technology. Hundreds of millions of gamers and creators each year seek out the best NVIDIA has to offer. 00:21:46.538 --> 00:21:53.241 At its core, computer graphics is about simulations - using mathematics and computer science to simulate the interactions of light 00:21:53.241 --> 00:22:00.052 and material, the physics of objects, particles, and waves; and now simulating intelligence and animation. 00:22:00.485 --> 00:22:07.859 The science, engineering, and artistry that we dedicate in pursuit of achieving mother nature's physics has led to incredible advances. 00:22:08.460 --> 00:22:14.266 And allowed our technology to contribute to advancing the basic sciences, the arts, and the industries. 00:22:15.067 --> 00:22:22.908 This last year, we introduced the 2nd generation of RTX - a new rendering approach that fuses rasterization and programmable shading with 00:22:22.908 --> 00:22:29.548 hardware-accelerated ray tracing and artificial intelligence. This is the culmination of ten years of research. 00:22:30.082 --> 00:22:37.255 RTX has reset computer graphics, giving developers a powerful new tool just as rasterization plateaus. 00:22:37.656 --> 00:22:44.129 Let me show you some amazing footage from games in development. The technology and artistry is amazing. 00:22:44.863 --> 00:22:48.333 We’re giving the world’s billion gamers an incredible reason to upgrade. 00:24:38.877 --> 00:24:41.513 RTX is a reset of computer graphics. 00:24:41.913 --> 00:24:49.187 It has enabled us to build Omniverse - a platform for connecting 3D worlds into a shared virtual world. 00:24:49.788 --> 00:24:57.295 Ones not unlike the science fiction metaverse first described by Neal Stephenson in his early 1990s novel, Snow Crash, where the metaverse 00:24:57.295 --> 00:25:05.103 would be collectives of shared 3D spaces and virtually-enhanced physical spaces that are extensions of the internet. 00:25:05.537 --> 00:25:14.934 Pieces of the early-metaverse vision are already here - massive online social games like Fortnite or user-created virtual worlds like Minecraft. 00:25:15.714 --> 00:25:22.187 Let me tell you about Omniverse from the perspective of two applications - design collaboration and digital twins. 00:25:22.954 --> 00:25:25.023 There are several major parts of the platform. 00:25:25.390 --> 00:25:30.272 First, the Omniverse Nucleus, a database engine that connects users 00:25:30.272 --> 00:25:34.332 and enables the interchange of 3D assets and scene descriptions. 00:25:34.699 --> 00:25:43.690 Once connected, designers doing modeling, layout, shading, animation, lighting, special effects or rendering can collaborate to create a scene. 00:25:44.075 --> 00:25:52.397 The Omniverse Nucleus is described with the open standard USD, Universal Scene Description, a fabulous interchange framework invented by Pixar. 00:25:53.251 --> 00:25:59.524 Multiple users can connect to Nucleus, transmitting and receiving changes to their world as USD snippets. 00:25:59.991 --> 00:26:06.798 The 2nd part of Omniverse is the composition, rendering, and animation engine - the simulation of the virtual world. 00:26:07.299 --> 00:26:15.774 Omniverse is a platform built from the ground up to be physically-based. It is fully path-traced. Physics is simulated with NVIDIA PhysX, 00:26:15.774 --> 00:26:21.713 materials are simulated with NVIDIA MDL and Omniverse is fully integrated with NVIDIA AI. 00:26:22.380 --> 00:26:30.255 Omniverse is cloud-native, multi-GPU scalable and runs on any RTX platform and streams remotely to any device. 00:26:31.022 --> 00:26:34.826 The third part is NVIDIA CloudXR, a stargate if you will. 00:26:35.060 --> 00:26:41.800 You can teleport into Omniverse with VR and AIs can teleport out of Omniverse with AR. 00:26:42.200 --> 00:26:44.469 Omniverse was released to open beta in December. 00:26:44.769 --> 00:26:46.638 Let me show you what talented creators are doing. 00:27:16.267 --> 00:27:18.903 Creators are doing amazing things with Omniverse. 00:27:19.237 --> 00:27:27.839 At Foster and Partners, designers in 17 locations around the world, are designing buildings together in their Omniverse shared virtual space. 00:27:28.480 --> 00:27:33.864 ILM is testing Omniverse to bring together internal and external tool pipelines from multiple studios. 00:27:33.864 --> 00:27:40.416 Omniverse lets them collaborate, render final shots in real time and create massive virtual sets like holodecks. 00:27:41.259 --> 00:27:48.400 Ericsson is using Omniverse to do real-time 5G wave propagation simulation, with many multi-path interferences. 00:27:48.800 --> 00:27:55.573 Twin Earth is creating a digital twin of Earth that will run on 20,000 NVIDIA GPUs. 00:27:55.573 --> 00:28:02.380 And Activision is using Omniverse to organize their more than 100,000 3D assets into a shared and searchable world. 00:28:12.691 --> 00:28:16.294 Bentley is the world's leading infrastructure engineering software company. 00:28:17.162 --> 00:28:23.566 Everything that's constructed - roads and bridges, rail and transit systems, airports and seaports - 00:28:23.566 --> 00:28:29.188 about 3% of the world's GDP or three-and-a-half trillion dollars a year. 00:28:29.207 --> 00:28:35.451 Bentley's software is used to design, model, and simulate the largest infrastructure projects in the world. 00:28:35.451 --> 00:28:39.129 90% of the world's top 250 engineering firms use Bentley. 00:28:40.018 --> 00:28:49.571 They have a new platform called iTwin - an exciting strategy to use the 3D model, after construction, to monitor and optimize the performance throughout its life. 00:28:50.295 --> 00:28:55.848 We are super excited to partner with Bentley to create infrastructure digital twins in Omniverse. 00:28:55.848 --> 00:29:01.614 Bentley is the first 3rd-party company to be developing a suite of applications on the Omniverse platform. 00:29:02.140 --> 00:29:08.546 This is just an awesome use of Omniverse, a great example of digital twins and Bentley is the perfect partner. 00:29:08.980 --> 00:29:15.120 And here's Perry Nightingale from WPP, the largest ad agency in the world, to tell you what they're doing. 00:29:18.490 --> 00:29:25.432 WPP is the largest marketing services organization on the planet, and because of that, we're also one of the largest production companies in the world 00:29:25.432 --> 00:29:28.594 That is a major carbon hotspot for us. 00:29:28.594 --> 00:29:35.138 We've partnered with NVIDIA, to capture locations virtually and bring them to life with studios in Omniverse. 00:29:35.138 --> 00:29:39.185 Over 10 billion points have turned into a giant mesh in Omniverse. 00:29:39.185 --> 00:29:45.361 For the first time, we can shoot locations virtually that are as real as the actual places themselves. 00:29:45.361 --> 00:29:48.838 Omniverse also changes the way we make work. 00:29:48.838 --> 00:29:56.691 A collaborative platform, that means multiple artists, at multiple points in the pipeline, in multiple parts of the world can collaborate on a single scene. 00:29:56.691 --> 00:30:04.968 Real time CGI in sustainable studios. Collaboration with Omniverse is the future of film at WPP. 00:30:07.505 --> 00:30:11.276 One of the most important features of Omniverse is that it obeys the laws of physics. 00:30:11.810 --> 00:30:16.347 Omniverse can simulate particles, fluids, materials, springs and cables. 00:30:16.881 --> 00:30:19.217 This is a fundamental capability for robotics. 00:30:19.617 --> 00:30:23.721 Once trained, the AI and software can be downloaded from Omniverse. 00:30:24.022 --> 00:30:30.735 In this video, you'll see Omniverse's physics simulation with rigid and soft bodies, fluids, and finite element modeling. 00:30:30.735 --> 00:30:32.652 And a lot more - enjoy! 00:32:16.601 --> 00:32:20.972 Omniverse is a physically-based virtual world where robots can learn to be robots. 00:32:21.372 --> 00:32:27.278 They'll come in all sizes and shapes - box movers, pick and place arms, forklifts, cars, trucks. 00:32:27.545 --> 00:32:34.886 In the future, a factory will be a robot, orchestrating many robots inside, building cars that are robots themselves. 00:32:35.320 --> 00:32:41.392 We can use Omniverse to create a virtual factory, train and simulate the factory and its robotic workers inside. 00:32:41.726 --> 00:32:48.766 The AI and software that run the virtual factory are exactly the same as what will run the actual one. 00:32:49.434 --> 00:32:53.538 The virtual and physical factories and their robots will operate in a loop. 00:32:54.339 --> 00:32:55.740 They are digital twins. 00:32:55.773 --> 00:32:59.944 Connecting to ERP systems, simulating the throughput of the factory, 00:33:00.111 --> 00:33:08.019 simulating new plant layouts, and becoming the dashboard of the operator - even uplinking into a robot to teleoperate it. 00:33:08.886 --> 00:33:13.291 BMW may very well be the world's largest custom-manufacturing company. 00:33:14.158 --> 00:33:20.665 BMW produces over 2 million cars a year. In their most advanced factory, a car a minute. Every car is different. 00:33:21.332 --> 00:33:28.806 We are working with BMW to create a future factory. Designed completely in digital, simulated from beginning to end in Omniverse, creating a 00:33:28.806 --> 00:33:33.277 digital twin, and operating a factory where robots and humans work together. 00:33:34.045 --> 00:33:36.714 Let's take a look at the BMW factory. 00:33:41.185 --> 00:33:50.962 Welcome to BMW Production, Jensen. I am pleased to show you why BMW sets the standards for innovation and flexibility. Our collaboration 00:33:50.962 --> 00:33:58.469 with NVIDIA Omniverse and NVIDIA AI leads into a new era of digitalization of automobile production. 00:33:58.936 --> 00:34:03.641 Fantastic to be with you, Milan. I am excited to do this virtual factory visit with you. 00:34:04.409 --> 00:34:14.218 We are inside the digital twin of BMW's assembly system, powered by Omniverse. For the first time, we are able to have our entire factory in 00:34:14.218 --> 00:34:24.629 simulation. Global teams can collaborate using different software packages like Revit, CATIA, or point clouds to design and plan the factory 00:34:24.662 --> 00:34:33.971 in real time 3D. The capability to operate in a perfect simulation revolutionizes BMW's planning processes. 00:34:38.509 --> 00:34:48.352 BMW regularly reconfigures its factories to accommodate new vehicle launches. Here we see 2 planning experts located in different parts of 00:34:48.352 --> 00:34:58.396 the world, testing a new line design in Omniverse. One of them wormholes into an assembly simulation with a motion capture suit, records 00:34:58.463 --> 00:35:04.202 task movements, while the other expert adjusts the line design - in real time. 00:35:04.202 --> 00:35:09.305 They work together to optimize the line as well as worker ergonomics and safety. 00:35:10.274 --> 00:35:12.310 “Can you tell how far I have to bend down there?” 00:35:12.443 --> 00:35:14.278 “First, I’ll get you a taller one” 00:35:14.645 --> 00:35:15.446 “Yeah, it’s perfect” 00:35:15.446 --> 00:35:19.450 We would like to be able to do this at scale in simulation. 00:35:20.218 --> 00:35:23.427 That's exactly why NVIDIA has Digital Human for simulation. 00:35:23.427 --> 00:35:26.861 Digital Humans are trained with data from real associates. 00:35:26.861 --> 00:35:32.623 You can then use Digital Humans in simulation to test new workflows for worker ergonomics and efficiency. 00:35:34.699 --> 00:35:41.353 Now, your factories employ 57,000 people that share workspace with many robots designed to make their jobs easier. 00:35:41.353 --> 00:35:42.039 Let's talk about them. 00:35:42.807 --> 00:35:43.791 You are right, Jensen. 00:35:43.791 --> 00:35:52.450 Robots are crucial for a modern production system. With NVIDIA Isaac robotics platform, BMW is deploying a fleet of 00:35:52.483 --> 00:35:57.506 intelligent robots for logistics to improve the material flow in our production. 00:35:57.506 --> 00:36:05.663 This agility is necessary since we produce 2.5 million vehicles per year, 99% of them are custom. 00:36:06.998 --> 00:36:12.654 Synthetic data generation and domain randomization available in Isaac are key to bootstrapping machine learning. 00:36:12.654 --> 00:36:18.993 Isaac Sim generates millions of synthetic images and vary the environment to teach the robots. 00:36:20.611 --> 00:36:29.731 Domain randomization can generate an infinite permutation of photorealistic objects, textures, orientations and lighting conditions. 00:36:29.731 --> 00:36:36.841 It is ideal for generating ground truth, whether for detection, segmentation, or depth perception. 00:36:38.763 --> 00:36:42.443 Let me show you an example of how we can combine it all to operate your factory. 00:36:42.443 --> 00:36:49.207 With NVIDIA's Fleet Command, your associates can securely orchestrate robots and other devices in the factory from Mission Control. 00:36:50.875 --> 00:37:02.258 They could monitor in real time complex manufacturing cells, update software over the air, launch robot missions and teleoperate. 00:37:02.258 --> 00:37:08.232 When a robot needs a helping hand, an alert can be sent to Mission Control and one of your associates can take control to help the robot. 00:37:11.362 --> 00:37:17.959 We're in the digital twin of one of your factories, but you have 30 others spread across 15 countries. 00:37:17.959 --> 00:37:20.142 The scale of BMW production is impressive, Milan. 00:37:20.972 --> 00:37:29.044 Indeed, Jensen, the scale and complexity of our production network requires BMW to constantly innovate. 00:37:29.044 --> 00:37:34.159 I am happy about the tight collaboration between our two companies. 00:37:34.159 --> 00:37:41.826 NVIDIA Omniverse and NVIDIA AI give us the chance to simulate all 31 factories in our production network. 00:37:41.826 --> 00:37:52.668 These new innovations will reduce the planning times, improve flexibility and precision and at the end, produce 30% more efficient planning processes. 00:37:53.804 --> 00:38:00.804 Milan, I could not be more proud of the innovations that our collaboration is bringing to the Factories of the Future. 00:38:00.804 --> 00:38:07.250 I appreciate you hosting me for a virtual visit of the digital twin of your BMW Production. 00:38:07.250 --> 00:38:08.719 It is a work of art! 00:38:13.224 --> 00:38:15.593 The ecosystem is really excited about Omniverse. 00:38:16.360 --> 00:38:23.434 This open platform, with USD universal 3D interchange, connects them into a large network of users. 00:38:24.135 --> 00:38:28.439 We have 12 connectors to major design tools already with another 40 in flight. 00:38:29.273 --> 00:38:31.942 Omniverse Connector SDK is available for download now. 00:38:32.810 --> 00:38:35.446 You can see that the most important design tools are already signed-up. 00:38:35.980 --> 00:38:43.220 Our lighthouse partners are from some of the world's largest industries - Media and Entertainment, Gaming, Architecture, Engineering, and 00:38:43.220 --> 00:38:48.459 Construction; Manufacturing, Telecommunications, Infrastructure, and Automotive. 00:38:48.959 --> 00:38:55.366 Computer makers worldwide are building NVIDIA-Certified workstations, notebooks, and servers optimized for Omniverse. 00:38:56.267 --> 00:39:00.237 And starting this summer, Omniverse will be available for enterprise license. 00:39:00.671 --> 00:39:06.544 Omniverse - NVIDIA's platform for creating and simulating shared virtual worlds. 00:39:08.446 --> 00:39:10.648 Data center is the new unit of computing. 00:39:11.482 --> 00:39:16.520 Cloud computing and AI are driving fundamental changes in the architecture of data centers. 00:39:17.221 --> 00:39:20.925 Traditionally, enterprise data centers ran monolithic software packages. 00:39:21.892 --> 00:39:29.845 Virtualization started the trend toward software-defined data centers - allowing applications to move about and letting IT manage from a "single-pane of glass". 00:39:30.368 --> 00:39:37.608 With virtualization, the compute, networking, storage, and security functions are emulated in software running on the CPU. 00:39:38.642 --> 00:39:46.717 Though easier to manage, the added CPU load reduced the data center's capacity to run applications, which is its primary purpose. 00:39:47.251 --> 00:39:51.522 This illustration shows the added CPU load in the gold-colored part of the stack. 00:39:51.922 --> 00:39:57.628 Cloud computing re-architected data centers again, now to provision services for billions of consumers. 00:39:58.229 --> 00:40:04.568 Monolithic applications were disaggregated into smaller microservices that can take advantage of any idle resource. 00:40:05.269 --> 00:40:10.207 Equally important, multiple engineering teams can work concurrently using CI/CD methods. 00:40:11.409 --> 00:40:17.314 Data center networks became swamped by east-west traffic generated by disaggregated microservices. 00:40:18.382 --> 00:40:22.119 CSPs tackled this with Mellanox's high-speed low-latency networking. 00:40:23.387 --> 00:40:31.295 Then, deep learning emerged. Magical internet services were rolled out, attracting more customers, and better engagement than ever. 00:40:32.663 --> 00:40:36.700 Deep learning is compute-intensive which drove adoption of GPUs. 00:40:37.368 --> 00:40:44.074 Nearly overnight, consumer AI services became the biggest users of GPU supercomputing technologies. 00:40:44.942 --> 00:40:52.416 Now, adding Zero-Trust security initiatives makes infrastructure software processing one of the largest workloads in the data center. 00:40:53.083 --> 00:41:00.090 The answer is a new type of chip for Data Center Infrastructure Processing like NVIDIA's Bluefield DPU. 00:41:00.090 --> 00:41:04.028 Let me illustrate this with our own cloud gaming service, GeForce Now, as an example. 00:41:04.428 --> 00:41:07.598 GeForce Now is NVIDIA's GeForce-in-the-cloud service. 00:41:08.232 --> 00:41:15.172 GeForce Now serves 10 million members in 70 countries. Incredible growth. 00:41:15.706 --> 00:41:19.417 GeForce Now is a seriously hard consumer service to deliver. 00:41:19.417 --> 00:41:32.475 Everything matters - speed of light, visual quality, frame rate, response, smoothness, start-up time, server cost, and most important of all, security. 00:41:33.390 --> 00:41:35.259 We're transitioning GeForce Now to BlueField. 00:41:35.893 --> 00:41:43.300 With Bluefield, we can isolate the infrastructure from the game instances, and offload and accelerate the networking, storage, and security. 00:41:44.068 --> 00:41:52.176 The GeForce Now infrastructure is costly. With BlueField, we will improve our quality of service and concurrent users at the same time - the 00:41:52.343 --> 00:41:53.944 ROI of BlueField is excellent. 00:41:54.712 --> 00:42:00.184 I'm thrilled to announce our first data center infrastructure SDK - DOCA 1.0 is available today! 00:42:00.818 --> 00:42:02.920 DOCA is our SDK to program BlueField. 00:42:03.587 --> 00:42:05.489 There's all kinds of great technology inside. 00:42:05.890 --> 00:42:15.375 Deep packet inspection, secure boot, TLS crypto offload, RegEX acceleration, and a very exciting capability - a hardware-based, 00:42:15.375 --> 00:42:20.813 real-time clock that can be used for synchronous data centers, 5G, and video broadcast. 00:42:21.705 --> 00:42:25.175 We have great partners working with us to optimize leading platforms on BlueField: 00:42:25.709 --> 00:42:34.752 Infrastructure software providers, edge and CDN providers, cybersecurity solutions, and storage providers - basically the world's leading 00:42:34.752 --> 00:42:36.754 companies in data center infrastructure. 00:42:37.421 --> 00:42:42.793 Though we're just getting started with BlueField 2, today we're announcing BlueField 3. 00:42:42.793 --> 00:42:44.461 22 billion transistors. 00:42:44.929 --> 00:42:47.698 The first 400 Gbps networking chip. 00:42:48.165 --> 00:42:54.638 16 Arm CPUs to run the entire virtualization software stack - for instance, running VMware ESX. 00:42:55.005 --> 00:43:03.614 BlueField 3 takes security to a whole new level, fully offloading and accelerating IPSEC and TLS cryptography, secret key management, and 00:43:03.614 --> 00:43:05.049 regular expression processing. 00:43:05.783 --> 00:43:09.253 We are on a pace to introduce a new Bluefield generation every 18 months. 00:43:10.120 --> 00:43:16.093 BlueField 3 will do 400 Gbps and be 10x the processing capability of BlueField 2. 00:43:17.161 --> 00:43:25.436 And BlueField 4 will do 800 Gbps and add NVIDIA's AI computing technologies to get another 10x boost. 00:43:26.236 --> 00:43:30.240 100x in 3 years -- and all of it will be needed. 00:43:31.175 --> 00:43:39.750 A simple way to think about this is that 1/3rd of the roughly 30 million data center servers shipped each year are consumed running the 00:43:39.750 --> 00:43:42.286 software-defined data center stack. 00:43:43.020 --> 00:43:46.090 This workload is increasing much faster than Moore's law. 00:43:46.690 --> 00:43:52.563 So, unless we offload and accelerate this workload, data centers will have fewer and fewer CPUs to run applications. 00:43:52.930 --> 00:43:54.264 The time for BlueField has come. 00:43:55.065 --> 00:44:04.241 At the beginning of the big bang of modern AI, we recognized the need to create a new kind of computer for a new way of developing software. 00:44:04.642 --> 00:44:08.545 Software will be written by software running on AI computers. 00:44:09.613 --> 00:44:19.723 This new type of computer will need new chips, new system architecture, new ways to network, new software, and new methodologies and tools. 00:44:20.524 --> 00:44:24.995 We've invested billions into this intuition, and it has proven helpful to the industry. 00:44:25.763 --> 00:44:29.767 It all comes together as DGX - a computer for AI. 00:44:30.234 --> 00:44:37.374 We offer DGX as a fully integrated system, as well as offer the components to the industry to create differentiated options. 00:44:37.841 --> 00:44:41.286 I am pleased to see so much AI research advancing because of DGX - 00:44:41.286 --> 00:44:48.997 top universities, research hospitals, telcos, banks, consumer products companies, car makers, and aerospace companies. 00:44:49.787 --> 00:44:56.319 DGX help their AI researchers - whose expertise is rare, scarce, and their work strategic. 00:44:56.319 --> 00:44:58.862 It is imperative to make sure they have the right instrument. 00:44:59.663 --> 00:45:06.570 Simply, if software is to be written by computers, then companies with the best software engineers will also need the best computers. 00:45:07.204 --> 00:45:10.441 We offer several configurations - all software compatible. 00:45:10.908 --> 00:45:18.849 The DGX A100 is a building block that contains 5 petaFLOPS of computing and superfast storage and networking to feed it. 00:45:19.316 --> 00:45:25.889 DGX Station is an AI data center in-a-box designed for workgroups - plugs into a normal outlet. 00:45:26.423 --> 00:45:35.634 And DGX SuperPOD is a fully integrated, fully network-optimized, AI-data-center-as-a-product. SuperPOD is for intensive AI research and development. 00:45:36.734 --> 00:45:41.083 NVIDIA's own new supercomputer, called Selene, is 4 SuperPODs. 00:45:41.083 --> 00:45:46.943 It is the 5th fastest supercomputer and the fastest industrial supercomputer in the world's Top 500. 00:45:47.745 --> 00:45:50.914 We have a new DGX Station 320G. 00:45:51.482 --> 00:46:04.591 DGX Station can train large models - 320 gigabytes of super-fast HBM2e connected to 4 A100 GPUs over 8 terabytes per second of memory bandwidth. 00:46:04.762 --> 00:46:07.998 8 terabytes transferred in one second. 00:46:08.332 --> 00:46:12.236 It would take 40 CPU servers to achieve this memory bandwidth. 00:46:13.303 --> 00:46:23.614 DGX Station plugs into a normal wall outlet, like a big gaming rig, consumes just 1,500 watts, and is liquid-cooled to a silent 37 db. 00:46:24.348 --> 00:46:27.284 Take a look at the cinematic that our engineers and creative team did. 00:47:30.814 --> 00:47:34.284 A CPU cluster of this performance would cost about a million dollars today. 00:47:34.818 --> 00:47:41.959 DGX Station is $149,000 - the ideal AI programming companion for every AI researcher. 00:47:42.659 --> 00:47:45.729 Today we are also announcing a new DGX SuperPOD. 00:47:46.063 --> 00:47:47.464 Three major upgrades: 00:47:48.098 --> 00:47:59.710 The new 80 gigabyte A100 which brings the SuperPOD to 90 terabytes of HBM2 memory, with aggregate bandwidth of 2.2 exabytes per second. 00:48:00.744 --> 00:48:10.921 It would take 11,000 CPU servers to achieve this bandwidth - about a 250-rack data center - 15 times bigger than the SuperPOD. 00:48:12.389 --> 00:48:15.993 Second, SuperPOD has been upgraded with NVIDIA BlueField-2. 00:48:16.727 --> 00:48:25.903 SuperPOD is now the world's first cloud-native supercomputer, multi-tenant shareable, with full isolation and bare-metal performance. 00:48:26.937 --> 00:48:33.610 And third, we're offering Base Command, the DGX management and orchestration tool used within NVIDIA. 00:48:34.845 --> 00:48:42.519 We use Base Command to support thousands of engineers, over two hundred teams, consuming a million-plus GPU-hours a week. 00:48:43.420 --> 00:48:48.258 DGX SuperPOD starts at seven million dollars and scales to sixty million dollars for a full system. 00:48:49.693 --> 00:48:51.862 Let me highlight 3 great uses of DGX. 00:48:52.562 --> 00:48:56.033 Transformers have led to dramatic breakthroughs in Natural Language Processing. 00:48:56.867 --> 00:49:02.539 Like RNN and LSTM, Transformers are designed to inference on sequential data. 00:49:03.173 --> 00:49:12.816 However, Transformers, more than meets the eyes, are not trained sequentially, but use a mechanism called attention such that Transformers 00:49:12.816 --> 00:49:14.117 can be trained in parallel. 00:49:14.751 --> 00:49:22.781 This breakthrough reduced training time, which more importantly enabled the training of huge models with a correspondingly enormous amount of data. 00:49:23.560 --> 00:49:27.864 Unsupervised learning can now achieve excellent results, but the models are huge. 00:49:28.665 --> 00:49:32.002 Google's Transformer was 65 million parameters. 00:49:32.869 --> 00:49:38.275 OpenAI's GPT-3 is 175 billion parameters. 00:49:38.742 --> 00:49:41.845 That's 3000x times larger in just 3 years. 00:49:42.579 --> 00:49:45.182 The applications for GPT-3 are really incredible. 00:49:45.649 --> 00:49:47.150 Generate document summaries. 00:49:47.718 --> 00:49:49.319 Email phrase completion. 00:49:50.120 --> 00:49:57.461 GPT-3 can even generate Javascript and HTML from plain English - essentially telling an AI to write code based on what you want it to do. 00:49:58.528 --> 00:50:03.734 Model sizes are growing exponentially - on a pace of doubling every two and half months. 00:50:04.534 --> 00:50:11.341 We expect to see multi-trillion-parameter models by next year, and 100 trillion+ parameter models by 2023. 00:50:13.110 --> 00:50:19.082 As a very loose comparison, the human brain has roughly 125 trillion synapses. 00:50:20.250 --> 00:50:22.252 So these transformer models are getting quite large. 00:50:22.986 --> 00:50:26.456 Training models of this scale is incredible computer science. 00:50:27.324 --> 00:50:32.996 Today, we are announcing NVIDIA Megatron - for training Transformers. 00:50:34.164 --> 00:50:41.505 Megatron trains giant Transformer models - it partitions and distributes the model for optimal multi-GPU and multi-node parallelism. 00:50:42.406 --> 00:50:51.481 Megatron does fast data loading, micro batching, scheduling and syncing, kernel fusing. It pushes the limits of every NVIDIA invention - 00:50:51.648 --> 00:50:56.186 NCCL, NVLink, Infiniband, Tensor Cores. 00:50:57.154 --> 00:51:01.691 Even with Megatron, a trillion-parameter model will take about 3-4 months to train on Selene. 00:51:02.659 --> 00:51:05.862 So, lots of DGX SuperPODS will be needed around the world 00:51:06.563 --> 00:51:10.834 Inferencing giant Transformer models is also a great computer science challenge. 00:51:11.234 --> 00:51:22.476 GPT-3 is so big, with so many floating-point operations, that it would take a dual-CPU server over a minute to respond to a single 128-word query. 00:51:23.713 --> 00:51:29.319 And GPT-3 is so large that it doesn't fit in GPU memory - so it will have to be distributed. 00:51:30.187 --> 00:51:33.423 multi-GPU multi-node inference has never been done. 00:51:34.257 --> 00:51:39.563 Today, we're announcing the Megatron Triton Inference Server. 00:51:40.764 --> 00:51:50.273 A DGX with Megatron Triton will respond within a second! Not a minute - a second! And for 16 queries at the same time. 00:51:51.174 --> 00:51:59.649 DGX is 1000 times faster and opens up many new use-cases, like call-center support, where a one-minute response is effectively unusable. 00:52:01.118 --> 00:52:03.019 Naver is Korea's #1 search engine. 00:52:03.920 --> 00:52:10.360 They installed a DGX SuperPOD and are running their AI platform CLOVA to train language models for Korean. 00:52:11.261 --> 00:52:15.275 I expect many leading service providers around the world to do the same 00:52:15.275 --> 00:52:20.455 Use DGX to develop and operate region-specific and industry-specific language services. 00:52:22.339 --> 00:52:31.114 NVIDIA Clara Discovery is our suite of acceleration libraries created for computational drug discovery - from imaging, to quantum chemistry, 00:52:31.748 --> 00:52:39.923 to gene variant-calling, to using NLP to understand genetics, and using AI to generate new drug compounds. 00:52:41.124 --> 00:52:44.528 Today we're announcing four new models available in Clara Discovery: 00:52:45.495 --> 00:52:49.831 MegaMolBART is a model for generating biomolecular compounds. 00:52:49.831 --> 00:52:56.617 This method has seen recent success with Insilico Medicine using AI to find a new drug in less than two years. 00:52:57.073 --> 00:53:05.549 NVIDIA ATAC-seq denoising algorithm for rare and single cell epi-genomics is helping to understand gene expression for individual cells. 00:53:06.516 --> 00:53:12.689 AlphaFold1 is a model that can predict the 3D structure of a protein from the amino acid sequence. 00:53:12.689 --> 00:53:18.895 GatorTron is the world's largest clinical language model that can read and understand doctors' notes. 00:53:18.895 --> 00:53:26.468 GatorTron was developed at UoF, using Megatron, and trained on the DGX SuperPOD gifted 00:53:26.468 --> 00:53:29.877 to his alma mater by Chris Malachowsky, who founded NVIDIA with Curtis and me. 00:53:31.041 --> 00:53:39.082 Oxford Nanopore is the 3rd generation genomics sequencing technology capable of ultra high throughput in digitizing biology - 00:53:39.082 --> 00:53:46.356 1/5 of the SARS-CoV-2 virus genomes in the global database were generated on Oxford Nanopore. 00:53:46.356 --> 00:53:52.596 Last year, Oxford Nanopore developed a diagnostic test for COVID-19 called LamPORE, which is used by NHS. 00:53:53.463 --> 00:53:55.865 Oxford Nanopore is GPU-accelerated throughout. 00:53:56.766 --> 00:54:06.539 DNA samples pass through nanopores and the current signal is fed into an AI model, like speech recognition, but trained to recognize genetic code. 00:54:07.577 --> 00:54:11.648 Another model called Medaka reads the code and detects genetic variants. 00:54:12.015 --> 00:54:14.784 Both models were trained on DGX SuperPOD. 00:54:15.352 --> 00:54:24.327 These new deep learning algorithms achieve 99.9% detection accuracy of single nucleotide variants - this is the gold standard of human sequencing. 00:54:24.861 --> 00:54:32.602 Pharma is a 1.3 trillion-dollar industry where a new drug can take 10+ years and fails 90% of the time. 00:54:33.770 --> 00:54:40.443 Schrodinger is the leading physics-based and machine learning computational platform for drug discovery and material science. 00:54:40.844 --> 00:54:50.808 Schrodinger is already a heavy user of NVIDIA GPUs, recently entering into an agreement to use hundreds of millions of NVIDIA GPU hours on the Google cloud. 00:54:51.554 --> 00:54:59.996 Some customers can't use the cloud, so today we are announcing a partnership to accelerate Schrodinger's drug discovery workflow with NVIDIA 00:55:00.063 --> 00:55:04.301 Clara Discovery libraries and NVIDIA DGX. 00:55:04.301 --> 00:55:10.740 The world's top 20 pharmas use Schrodinger today. Their researchers are going to see a giant boost in productivity. 00:55:11.574 --> 00:55:19.182 Recursion is a biotech company using leading-edge computer science to decode biology to industrialize drug discovery. 00:55:19.949 --> 00:55:30.506 The Recursion Operating System is built on NVIDIA DGX SuperPOD for generating, analyzing and gaining insight from massive biological and chemical datasets. 00:55:31.561 --> 00:55:37.767 They call their SuperPOD the BioHive-1 - it's the most powerful computer at any pharma today. 00:55:38.702 --> 00:55:46.409 Using deep learning on DGX, Recursion is classifying cell responses after exposure to small molecule drugs. 00:55:47.277 --> 00:55:56.226 Quantum computing is a field of physics that studies the use of natural quantum behavior - superposition, entanglement, and interference - to build a computer. 00:55:56.853 --> 00:56:01.491 The computation is performed using quantum circuits that operate on quantum bits - called qubits. 00:56:02.492 --> 00:56:09.866 Qubits can be 0 or 1, like a classical computing bit, but also in superposition - meaning they exist simultaneously in both states. 00:56:10.533 --> 00:56:15.672 The qubits can be entangled where the behavior of one can affect or control the behavior of others. 00:56:16.206 --> 00:56:22.078 Adding and entangling more qubits lets quantum computers calculate exponentially more information. 00:56:22.812 --> 00:56:26.883 There is a large community around the world doing research in quantum computers and algorithms. 00:56:27.550 --> 00:56:32.689 Well over 50 teams in industry, academia, and national labs are researching the field. 00:56:33.390 --> 00:56:34.624 We're working with many of them. 00:56:35.325 --> 00:56:44.067 Quantum computing can solve Exponential Order Complexity problems, like factoring large numbers for cryptography, simulating atoms and 00:56:44.067 --> 00:56:49.806 molecules for drug discovery, finding shortest path optimizations, like the traveling salesman problem, 00:56:50.507 --> 00:56:57.647 The limiter in quantum computing is decoherence, falling out of quantum states, caused by the tiniest of background noise 00:56:58.081 --> 00:56:59.649 So error correction is essential. 00:57:00.483 --> 00:57:07.323 It is estimated that to solve meaningful problems, several million physical qubits will be required to sufficiently error correct. 00:57:07.957 --> 00:57:17.357 The research community is making fast progress, doubling physical Qubits each year, so likely achieving the milestone by 2035 to 2040. 00:57:17.357 --> 00:57:19.759 Well within my career horizon. 00:57:20.036 --> 00:57:26.843 In the meantime, our mission is to help the community research the computer of tomorrow with the fastest computer of today. 00:57:27.777 --> 00:57:34.117 Today, we're announcing cuQuantum - an acceleration library designed for simulating quantum circuits 00:57:34.484 --> 00:57:37.520 for both Tensor Network Solvers and State Vector Solvers. 00:57:38.788 --> 00:57:44.260 It is optimized to scale to large GPU memories, multiple GPUs, and multiple DGX nodes. 00:57:45.428 --> 00:57:54.571 The speed-up of cuQuantum on DGX is excellent. Running the cuQuantum Benchmark, state vector simulation takes 10 days on a dual-CPU server 00:57:55.271 --> 00:58:02.979 but only 2 hours on a DGX A100. cuQuantum on DGX can productively simulate 10's of qubits. 00:58:04.481 --> 00:58:15.792 And Caltech, using Contengra/Quimb, simulated the Sycamore quantum circuit at depth 20 in record time using cuQuantum on NVIDIA's Selene supercomputer. 00:58:16.392 --> 00:58:23.299 What would have taken years on CPUs can now run in a few days on cuQuantum and DGX. 00:58:24.167 --> 00:58:31.808 cuQuantum will accelerate quantum circuit simulators so researchers can design better quantum computers and verify their results, architect 00:58:31.808 --> 00:58:38.348 hybrid quantum-classical systems -and discover more Quantum-Optimal algorithms like Shor's and Grover's. 00:58:39.115 --> 00:58:43.853 cuQuantum on DGX is going to give the quantum community a huge boost. 00:58:44.554 --> 00:58:49.659 I'm hoping cuQuantum will do for quantum computing what cuDNN did for deep learning. 00:58:50.493 --> 00:58:54.964 Modern data centers host diverse applications that require varying system architectures. 00:58:55.532 --> 00:59:01.337 Enterprise servers are optimized for a balance of strong single-threaded performance and a nominal number of cores. 00:59:01.905 --> 00:59:10.046 Hyperscale servers, optimized for microservice containers, are designed for a high number of cores, low cost, and great energy-efficiency. 00:59:10.680 --> 00:59:14.918 Storage servers are optimized for large number of cores and high IO throughput. 00:59:15.285 --> 00:59:22.592 Deep learning training servers are built like supercomputers - with the largest number of fast CPU cores, the fastest memory, 00:59:22.592 --> 00:59:26.796 the fastest IO, and high-speed links to connect the GPUs. 00:59:26.996 --> 00:59:33.670 Deep learning inference servers are optimized for energy-efficiency and best ability to process a large number of models concurrently. 00:59:34.437 --> 00:59:45.448 The genius of the x86 server architecture is the ability to do a good job using varying configurations of the CPU, memory, PCI express, and 00:59:45.448 --> 00:59:48.384 peripherals to serve all of these applications. 00:59:49.218 --> 00:59:56.793 Yet processing large amounts of data remains a challenge for computer systems today - this is particularly true for AI models like 00:59:56.793 --> 00:59:58.595 transformers and recommender systems. 00:59:59.228 --> 01:00:03.099 Let me illustrate the bottleneck with half of a DGX. 01:00:03.733 --> 01:00:10.440 Each Ampere GPU is connected to 80GB of super fast memory running at 2 TB/sec. 01:00:11.374 --> 01:00:18.715 Together, the 4 Amperes process 320 GB at 8 Terabytes per second. 01:00:19.382 --> 01:00:27.308 Contrast that with CPU memory, which is 1TB large, but only 0.2 Terabytes per second. 01:00:27.308 --> 01:00:32.178 The CPU memory is 3 times larger but 40 times slower than the GPU. 01:00:33.029 --> 01:00:39.102 We would love to utilize the full 1,320 GB of memory of this node to train AI models. 01:00:39.869 --> 01:00:41.404 So, why not something like this? 01:00:42.171 --> 01:00:49.379 Make faster CPU memories, connect 4 channels to the CPU, a dedicated channel to feed each GPU. 01:00:49.846 --> 01:00:53.750 Even if a package can be made, PCIe is now the bottleneck. 01:00:54.450 --> 01:01:02.258 We can surely use NVLINK. NVLINK is fast enough. But no x86 CPU has NVLINK, not to mention 4 NVLINKS. 01:01:03.593 --> 01:01:13.469 Today, we're announcing our first data center CPU, Project Grace, named after Grace Hopper, a computer scientist and U.S. Navy rear Admiral, 01:01:13.903 --> 01:01:16.105 who in the '50s pioneered computer programming. 01:01:17.006 --> 01:01:23.646 Grace is Arm-based and purpose-built for accelerated computing applications of large amounts of data - such as AI. 01:01:24.380 --> 01:01:26.358 Grace highlights the beauty of Arm. 01:01:26.358 --> 01:01:33.527 Their IP model allowed us to create the optimal CPU for this application which achieves x-factors speed-up. 01:01:34.323 --> 01:01:45.568 The Arm core in Grace is a next generation off-the-shelf IP for servers. Each CPU will deliver over 300 SpecInt with a total of over 2,400 01:01:45.568 --> 01:01:51.774 SPECint_rate CPU performance for an 8-GPU DGX. 01:01:52.275 --> 01:02:00.149 For comparison, todays DGX, the highest performance computer in the world today is 450 SPECint_rate. 01:02:00.883 --> 01:02:07.256 2400 SPECint_rate with Grace versus 450 SPECint_rate today. 01:02:07.890 --> 01:02:17.133 So look at this again - Before, After, Before, After. 01:02:18.367 --> 01:02:21.604 Amazing increase in system and memory bandwidth. 01:02:22.538 --> 01:02:25.174 Today, we're introducing a new kind of computer. 01:02:26.075 --> 01:02:28.578 The basic building block of the modern data center. 01:02:29.479 --> 01:02:30.146 Here it is. 01:02:42.391 --> 01:02:50.688 What I'm about to show you brings together the latest GPU accelerated computing, Mellanox high performance networking, and something brand new. 01:02:51.267 --> 01:02:52.935 The final piece of the puzzle. 01:02:59.142 --> 01:03:07.416 The world's first CPU designed for terabyte-scale accelerated computing... her secret codename - GRACE. 01:03:09.385 --> 01:03:19.310 This powerful, Arm-based CPU gives us the third foundational technology for computing, and the ability to rearchitect every aspect of the data center for AI. 01:03:20.663 --> 01:03:27.804 We're thrilled to announce the Swiss National Supercomputing Center will build a supercomputer powered by Grace and our next generation GPU. 01:03:28.504 --> 01:03:39.582 This new supercomputer, called Alps, will be 20 exaflops for AI, 10 times faster than the world's fastest supercomputer today. 01:03:40.583 --> 01:03:48.100 Alps will be used to do whole-earth-scale weather and climate simulation, quantum chemistry and quantum physics for the Large Hadron Collider. 01:03:48.925 --> 01:03:54.730 Alps will be built by HPE and is come on-line in 2023. 01:03:54.730 --> 01:04:02.605 We're thrilled by the enthusiasm of the supercomputing community, welcoming us to make Arm a top-notch scientific computing platform. 01:04:03.372 --> 01:04:11.214 Our data center roadmap is now a rhythm consisting of 3-chips: CPU, GPU, and DPU. 01:04:12.114 --> 01:04:16.986 Each chip architecture has a two-year rhythm with likely a kicker in between. 01:04:17.954 --> 01:04:20.590 One year will focus on x86 platforms. 01:04:21.023 --> 01:04:23.125 One year will focus on Arm platforms. 01:04:23.826 --> 01:04:26.662 Every year will see new exciting products from us. 01:04:27.396 --> 01:04:34.403 The NVIDIA architecture and platforms will support x86 and Arm - whatever customers and markets prefer. 01:04:35.338 --> 01:04:39.075 Three chips. Yearly Leaps. One Architecture. 01:04:39.976 --> 01:04:42.311 Arm is the most popular CPU in the world. 01:04:43.246 --> 01:04:51.921 For good reason - its super energy-efficient. Its open licensing model inspires a world of innovators to create products around it. 01:04:52.955 --> 01:04:55.391 Arm is used broadly in mobile and embedded today. 01:04:56.392 --> 01:05:03.018 For other markets like the cloud, enterprise and edge data centers, supercomputing and PCs. 01:05:03.018 --> 01:05:07.096 Arm is just starting and has great growth opportunities. 01:05:07.637 --> 01:05:14.110 Each market has different applications and has unique systems, software, peripherals, and ecosystems. 01:05:15.011 --> 01:05:18.347 For the markets we serve, we can accelerate Arm's adoption. 01:05:19.248 --> 01:05:20.950 Let's start with the big one - Cloud. 01:05:21.951 --> 01:05:29.558 One of the earliest designers of Arm CPUs for data centers is AWS - its Graviton CPUs are extremely impressive. 01:05:30.259 --> 01:05:37.767 Today, we're announcing NVIDIA and AWS are partnering to bring Graviton2 and NVIDIA GPUs together. 01:05:38.668 --> 01:05:44.774 This partnership brings Arm into the most demanding cloud workloads - AI and cloud gaming. 01:05:45.408 --> 01:05:50.079 Mobile gaming is growing fast and is the primary form of gaming in some markets. 01:05:50.813 --> 01:05:58.621 With AWS-designed Graviton2, users can stream Arm-based applications and Android games straight from AWS. 01:05:59.422 --> 01:06:00.656 It's expected later this year. 01:06:01.524 --> 01:06:08.197 We are announcing a partnership with Ampere Computing to create a scientific and cloud computing SDK and reference system. 01:06:09.265 --> 01:06:19.141 Ampere Computing's Altra CPU is excellent - 80 cores, 285 SPECint17, right up there with the highest performance x86. 01:06:20.176 --> 01:06:25.581 We are seeing excellent reception at supercomputing centers around the world and at Android cloud gaming services. 01:06:26.282 --> 01:06:33.689 We are also announcing a partnership with Marvell to create an edge and enterprise computing SDK and reference system. 01:06:34.623 --> 01:06:42.765 Marvell Octeon excels at IO, storage and 5G processing. This system is ideal for hyperconverged edge servers. 01:06:44.066 --> 01:06:50.439 We're announcing a partnership with Mediatek to create a reference system and SDK for Chrome OS and Linux PC's. 01:06:51.140 --> 01:06:54.777 Mediatek is the world's largest SOC maker. 01:06:55.611 --> 01:07:01.250 Combining NVIDIA GPUs and Mediatek SOCs will make excellent PCs and notebooks. 01:07:02.485 --> 01:07:09.492 AI, computers automating intelligence, is the most powerful technology force of our time. 01:07:10.192 --> 01:07:11.927 We see AI in four waves. 01:07:12.261 --> 01:07:20.202 The first wave was to reinvent computing for this new way of doing software - we're all in and have been driving this for nearly 10 years. 01:07:20.903 --> 01:07:27.810 The first adopters of AI were the internet companies - they have excellent computer scientists, large computing infrastructures, and the 01:07:27.810 --> 01:07:29.712 ability to collect a lot of training data. 01:07:30.880 --> 01:07:32.915 We are now at the beginning of the next wave. 01:07:33.716 --> 01:07:42.691 The next wave is enterprise and the industrial edge, where AI can revolutionize the world's largest industries. From manufacturing, 01:07:42.792 --> 01:07:47.797 logistics, agriculture, healthcare, financial services, and transportation. 01:07:48.531 --> 01:07:53.869 There are many challenges to overcome, one of which is connectivity, which 5G will solve. 01:07:54.503 --> 01:08:01.510 And then autonomous systems. Self-driving cars are an excellent example. But everything that moves will eventually be autonomous. 01:08:02.244 --> 01:08:09.518 The industrial edge and autonomous systems are the most challenging, but also the largest opportunities for AI to make an impact. 01:08:10.152 --> 01:08:17.560 Trillion dollar industries can soon apply AI to improve productivity, and invent new products, services and business models. 01:08:18.761 --> 01:08:24.800 We have to make AI easier to use - turn AI from computer science to computer products. 01:08:25.201 --> 01:08:32.842 We're building the new computing platform for this fundamentally new software approach - the computer for the age of AI. 01:08:33.676 --> 01:08:47.103 AI is not just about an algorithm - building and operating AI is a fundamental change in every aspect of software - Andrej Karpathy rightly called it Software 2.0. 01:08:47.123 --> 01:08:54.663 Machine learning, at the highest level, is a continuous learning system that starts with data scientists developing data strategies 01:08:54.663 --> 01:09:00.636 and engineering predictive features - this data is the digital life experience of a company. 01:09:01.470 --> 01:09:06.675 Training involves inventing or adapting an AI model that learns to make the desired predictions. 01:09:07.543 --> 01:09:14.049 Simulation and validation test the AI application for accuracy, generalization, and potential bias. 01:09:14.550 --> 01:09:22.165 And finally, orchestrating a fleet of computers, whether in your data center or at the edge in the warehouse, farms, or wireless base stations. 01:09:22.191 --> 01:09:31.233 NVIDIA created the chips, systems, and libraries needed for end-to-end machine learning - for example, technologies like Tensor Core GPUs, 01:09:31.333 --> 01:09:36.972 NVLINK, DGX, cuDNN, RAPIDS, NCCL, GPU Direct, DOCA, and so much more. 01:09:37.673 --> 01:09:39.642 We call the platform NVIDIA AI. 01:09:40.709 --> 01:09:46.649 NVIDIA AI libraries accelerate every step, from data processing to fleet orchestration. 01:09:47.416 --> 01:09:51.787 NVIDIA AI is integrated into all of the industry's popular tools and workflows. 01:09:52.788 --> 01:10:00.196 NVIDIA AI is in every cloud, used by the world's largest companies, and by over 7,500 AI startups around the world. 01:10:00.596 --> 01:10:11.006 And NVIDIA AI runs on any system that includes NVIDIA GPUs, from PCs and laptops, to workstations, to supercomputers, in any cloud, to our 01:10:11.006 --> 01:10:13.209 $99 Jetson robot computer. 01:10:13.976 --> 01:10:18.247 One segment of computing we've not served is enterprise computing. 01:10:19.181 --> 01:10:23.118 70% of the world's enterprises run VMware, as we do at NVIDIA. 01:10:23.619 --> 01:10:27.890 VMware was created to run many applications on one virtualized machine. 01:10:28.857 --> 01:10:36.098 AI, on the other hand, runs a single job, bare-metal, on multiple GPUs and often multiple nodes. 01:10:36.932 --> 01:10:44.473 All of the NVIDIA optimizations for compute and data transfer are now plumbed through the VMware stack so AI workloads can be distributed to 01:10:44.473 --> 01:10:47.610 multiple systems and achieve bare-metal performance. 01:10:48.444 --> 01:10:51.914 The VMware stack is also offloaded and accelerated on NVIDIA BlueField. 01:10:52.815 --> 01:11:02.224 NVIDIA AI now runs in its full glory on VMware, which means everything that has been accelerated by NVIDIA AI now runs great on VMware. 01:11:03.158 --> 01:11:07.630 AI applications can be deployed and orchestrated with Kubernetes running on VMware Tanzu. 01:11:08.364 --> 01:11:11.867 We call this platform NVIDIA EGX for Enterprise. 01:11:12.534 --> 01:11:22.444 The enterprise IT ecosystem is thrilled - finally the 300,000 VMware enterprise customers can easily build an AI computing infrastructure 01:11:22.778 --> 01:11:25.814 that seamlessly integrates into their existing environment. 01:11:26.548 --> 01:11:32.721 In total, over 50 servers from the world's top server makers will be certified for NVIDIA EGX Enterprise. 01:11:33.022 --> 01:11:38.227 BlueField 2 offloads and accelerates the VMware stack and does the networking for distributed computing. 01:11:38.794 --> 01:11:46.001 Enterprise can choose big or small GPUs for heavy-compute or heavy-graphics workloads like Omniverse, or mix and match. 01:11:46.902 --> 01:11:48.370 All run NVIDIA AI. 01:11:49.171 --> 01:11:58.013 Enterprise companies make up the world's largest industries and they operate at the edge - in hospitals, factories, plants, warehouses, 01:11:58.013 --> 01:12:02.117 stores, farms, cities and roads - far from data centers. 01:12:03.185 --> 01:12:04.787 The missing link is 5G. 01:12:05.621 --> 01:12:10.059 Consumer 5G is great, but Private 5G is revolutionary. 01:12:10.893 --> 01:12:20.102 Today, we're announcing the Aerial A100 - bringing together 5G and AI into a new type of computing platform designed for the edge. 01:12:20.636 --> 01:12:29.578 Aerial A100 integrates the Ampere GPU and BlueField DPU into one card - this is the most advanced PCI express card ever created. 01:12:29.978 --> 01:12:35.918 So, it's not a surprise that Aerial A100 in an EGX system will be a complete 5G base station. 01:12:36.952 --> 01:12:52.201 Aerial A100 delivers up to full 20 Gbps and can process up to 9 100Mhz massive MIMO for 64T64R - or 64 transmit and 64 receive antenna arrays 01:12:52.201 --> 01:12:54.603 - state of the art capabilities. 01:12:55.270 --> 01:13:06.843 Aerial A100 is software-defined, with accelerated features like PHY, Virtual Network Functions, network acceleration, packet pacing, and line-rate cryptography. 01:13:08.117 --> 01:13:16.825 Our partners ERICSSON, Fujitsu, Mavenir, Altran, and Radisys will build their total 5G solutions on top of the Aerial library. 01:13:17.826 --> 01:13:28.303 NVIDIA EGX server with Aerial A100 is the first 5G base-station that is also a cloud-native, secure, AI edge data center. 01:13:29.138 --> 01:13:32.708 We have brought the power of the cloud to the 5G edge. 01:13:33.609 --> 01:13:36.779 Aerial also extends the power of 5G into the cloud. 01:13:37.312 --> 01:13:43.152 Today, we are excited to announce that Google will support NVIDIA Aerial in the GCP cloud. 01:13:43.986 --> 01:13:45.621 I have an important new platform to tell you about. 01:13:46.522 --> 01:13:54.229 The rise of microservice-based applications and hybrid-cloud has exposed billions of connections in a data center to potential attack. 01:13:54.863 --> 01:14:04.889 Modern Zero-Trust security models assume the intruder is already inside and all container-to-container communications should be inspected, even within a node. 01:14:05.274 --> 01:14:06.708 This is not possible today. 01:14:07.476 --> 01:14:12.448 The CPU load of monitoring every piece of traffic is simply too great. 01:14:13.215 --> 01:14:20.489 Today, we are announcing NVIDIA Morpheus - a data center security platform for real-time all-packet inspection. 01:14:21.156 --> 01:14:28.530 Morpheus is built on NVIDIA AI, NVIDIA BlueField, Net-Q network telemetry software, and EGX. 01:14:29.431 --> 01:14:39.374 We're working to create solutions with industry leaders in data center security - Fortinet, Red Hat, Cloudflare, Splunk, F5, and Aria 01:14:39.374 --> 01:14:46.782 Cybersecurity. And early customers - Booz Allen Hamilton, Best Buy, and of course, our own team at NVIDIA. 01:14:47.716 --> 01:14:49.952 Let me show you how we're using Morpheus at NVIDIA. 01:14:52.654 --> 01:14:54.089 It starts with a network. 01:14:54.523 --> 01:15:01.563 Here we see a representation of a network, where dots are servers and lines (the edges) are packets flowing between those servers. 01:15:01.897 --> 01:15:09.938 Except in this network, Morpheus is deployed. This enables AI inferencing across your entire network, including east/west traffic. The 01:15:09.938 --> 01:15:17.613 particular model being used here has been trained to identify sensitive information - AWS credentials, GitHub credentials, private keys, 01:15:17.713 --> 01:15:22.751 passwords. If observed in the packet, these would appear as red lines, and we don't see any of that. 01:15:23.519 --> 01:15:24.820 Uh oh, what happened. 01:15:25.454 --> 01:15:29.391 An updated configuration was deployed to a critical business app on this server. 01:15:30.025 --> 01:15:34.830 This update accidentally removed encryption, and now everything that communicates with that app 01:15:34.830 --> 01:15:37.933 sends and receives sensitive credentials in the clear. 01:15:38.800 --> 01:15:47.075 This can quickly impact additional servers. This translates to continuing exposure on the network. The AI model in Morpheus is searching 01:15:47.075 --> 01:15:53.982 through every packet for any of these credentials, continually flagging when it encounters such data. And rather than using pattern 01:15:53.982 --> 01:16:00.589 matching, this is done with a deep neural network - trained to generalize and identify patterns beyond static rule sets. 01:16:01.557 --> 01:16:08.330 Notice all of the individual lines. It's easy to see how quickly a human could be overwhelmed by the vast amount of data coming in. 01:16:09.364 --> 01:16:13.835 Scrolling through the raw data gives a sense of the massive scale and complexity that is involved. 01:16:14.636 --> 01:16:19.408 With Morpheus, we immediately see the lines that represent leaked sensitive information. 01:16:20.008 --> 01:16:27.149 By hovering over one of those red lines, we show complete info about the credential, making it easy to triage and remediate. 01:16:27.950 --> 01:16:35.991 But what happens when this remediation is necessary? Morpheus enables cyber applications to integrate and collect information for automated 01:16:36.058 --> 01:16:38.894 incident management and action prioritization. 01:16:39.361 --> 01:16:48.003 Originating servers, destination servers, actual exposed credentials, and even the raw data is available. This speeds recovery and informs 01:16:48.036 --> 01:16:50.739 which keys were compromised and need be rotated. 01:16:51.206 --> 01:16:53.976 With Morpheus, the chaos becomes manageable. 01:16:56.812 --> 01:17:02.951 The IT ecosystem has been hungry for an AI computing platform that is enterprise and edge ready. 01:17:03.619 --> 01:17:10.258 NVIDIA AI on EGX with Aerial 5G is the foundation of what the IT ecosystem has been waiting for. 01:17:10.859 --> 01:17:20.535 We are supported by leaders from all across the IT industry - from systems, infrastructure software, storage and security, data analytics, 01:17:20.969 --> 01:17:26.708 industrial edge solutions, manufacturing design and automation, to 5G infrastructure. 01:17:27.609 --> 01:17:36.018 To complete our enterprise offering, we now have NVIDIA AI Enterprise software so businesses can get direct-line support from NVIDIA. 01:17:37.085 --> 01:17:43.992 NVIDIA AI Enterprise is optimized and certified for VMware, and offers services and support needed by mission-critical enterprises. 01:17:45.394 --> 01:17:49.064 Deep learning has unquestionably revolutionized computing. 01:17:49.798 --> 01:17:54.002 Researchers continue to innovate at lightspeed with new models and new variants. 01:17:54.536 --> 01:18:01.109 We're creating new systems to expand AI into new segments - like Arm, enterprise, or 5G edge. 01:18:01.443 --> 01:18:06.615 We're also using these systems to do basic research in AI and building new AI products and services. 01:18:07.249 --> 01:18:11.186 Let me show you some of our work in AI, basic and applied research. 01:18:11.853 --> 01:18:14.189 DLSS: Deep Learning Super Sampling. 01:18:14.923 --> 01:18:18.060 StyleGAN: AI high resolution image generator. 01:18:18.960 --> 01:18:24.399 GANcraft: a neural rendering engine, turning Minecraft into realistic 3D. 01:18:25.333 --> 01:18:30.005 GANverse3D: turns photographs into animatable 3D models. 01:18:31.106 --> 01:18:38.613 Face Vid2Vid: a talking-head rendering engine that can reduce streaming bandwidth by 10x, while re-posing the head and eyes. 01:18:39.448 --> 01:18:46.421 Sim2Real: a quadruped trained in Omniverse, and the AI can run in a real robot - digital twins. 01:18:47.255 --> 01:18:52.527 SimNet: a Physics Informed Neural Network solver that can simulate large-scale multi-physics. 01:18:53.295 --> 01:18:56.698 BioMegatron: the largest biomedical language model ever trained. 01:18:57.866 --> 01:19:00.836 3DGT: Omniverse synthetic data generation. 01:19:01.670 --> 01:19:04.940 and OrbNet: a machine learning quantum solver for quantum chemistry. 01:19:05.640 --> 01:19:09.277 This is just a small sampling of the AI work we're doing at NVIDIA. 01:19:09.778 --> 01:19:12.581 We're building AIs to use in our products and platforms. 01:19:13.381 --> 01:19:17.686 We're also packaging up the models to be easily integrated into your applications. 01:19:18.353 --> 01:19:22.491 These are essentially no-coding open-source applications that you can modify. 01:19:23.225 --> 01:19:28.330 Now, we're offering NGC pre-trained models, that you can plug into these applications, or ones you develop. 01:19:28.897 --> 01:19:34.669 These pre-trained models are production quality, trained by experts, and will continue to benefit from refinement. 01:19:35.403 --> 01:19:39.574 There are new credentials to tell you about the models' development, testing, and use. 01:19:40.208 --> 01:19:42.944 And each comes with a reference application sample code. 01:19:43.745 --> 01:19:52.731 NVIDIA pre-trained models are state-of-the-art and meticulously trained, but there is infinite diversity of application domains, environments, and specializations. 01:19:53.522 --> 01:19:58.527 No one has all the data - sometimes it's rare, sometimes they're trade-secrets. 01:19:59.194 --> 01:20:05.167 So we created technology for you to fine-tune and adapt NVIDIA pre-trained models for your applications. 01:20:06.001 --> 01:20:10.705 TAO applies transfer learning on your data to fine-tune our models with your data. 01:20:11.173 --> 01:20:19.314 TAO has an excellent federated learning system to let multiple parties collectively train a shared model while protecting data privacy. 01:20:20.081 --> 01:20:27.255 NVIDIA federated learning is a big deal - researchers at different hospitals can collaborate on one AI model while keeping their data 01:20:27.255 --> 01:20:29.291 separate to protect patient privacy. 01:20:30.025 --> 01:20:35.330 TAO uses NVIDIA's great TensorRT to optimize the model for the target GPU system. 01:20:35.964 --> 01:20:42.437 With NVIDIA pre-trained models and TAO, something previously impossible for many, can now be done in hours. 01:20:43.171 --> 01:20:56.748 Fleet Command is a cloud-native platform for securely operating and orchestrating AI across a distributed fleet of computers - it was purpose-built for operating AI at the edge. 01:20:57.686 --> 01:21:06.828 Fleet Command running on Certified EGX systems will feature secure boot, attestation, uplink and downlink, to a confidential enclave. 01:21:07.762 --> 01:21:11.032 From any cloud or on-prem, you can monitor the health of your fleet. 01:21:11.867 --> 01:21:17.320 Let's take a look at how one customer is using NGC pre-trained models and TAO to fine-tune models 01:21:17.320 --> 01:21:22.969 and run in our Metropolis smart city application, orchestrated by Fleet Command. 01:21:25.213 --> 01:21:28.884 No two industrial environments are the same. And conditions routinely change 01:21:29.284 --> 01:21:36.258 Adapting, managing, and operationalizing AI-enabled applications at the edge for unique, specific sites can be incredibly challenging - 01:21:36.258 --> 01:21:38.560 requiring a lot of data and time for model training. 01:21:39.394 --> 01:21:42.898 Here, we're building applications for a factory with multiple problems to solve 01:21:43.632 --> 01:21:46.401 Understanding how a factory floor space is used over time 01:21:46.801 --> 01:21:50.438 Ensuring worker safety, with constantly evolving machinery and risk factors 01:21:51.106 --> 01:21:54.309 Inspecting products on the factory line, where operations change routinely 01:21:55.277 --> 01:22:00.982 We start with a Metropolis application running 3 pretrained models from NGC. We are up and running in minutes. 01:22:01.216 --> 01:22:07.555 But since the environment is visually quite different from the data used to train this model, the video analytics accuracy at this specific 01:22:07.555 --> 01:22:12.527 site isn't great. People are not being accurately recognized and tracked and defects are being missed. 01:22:13.061 --> 01:22:15.163 Now let's use NVIDIA TAO to solve for this. 01:22:15.430 --> 01:22:22.037 With this simple UI, we re-train and adapt our pre-trained models with labelled data from the specific environment we're deploying in. 01:22:22.637 --> 01:22:23.838 We select our datasets. 01:22:24.439 --> 01:22:29.978 Each is a few 100 images, as opposed to millions of labelled images required if we were to train this from scratch. 01:22:30.645 --> 01:22:34.616 With NVIDIA Tao we go from 65% to over 90% accuracy. 01:22:35.216 --> 01:22:40.922 And through pruning & quantization, compute complexity is reduced by 2x - with no reduction in accuracy 01:22:40.922 --> 01:22:42.557 and real-time performance is maintained. 01:22:44.159 --> 01:22:48.863 In this example, the result is 3 models specifically trained for our factory - all in just minutes 01:22:49.731 --> 01:22:55.003 With one-click, we update and deploy these optimized models onto NVIDIA Certified servers with Fleet Command, 01:22:55.236 --> 01:22:57.305 seamlessly and securely from the cloud. 01:22:58.573 --> 01:23:05.814 From secure boot to our confidential AI Enclave in GPUs, application data and critical intellectual property remains safe. 01:23:06.314 --> 01:23:09.851 AI accuracy, system performance and health can be monitored remotely. 01:23:10.385 --> 01:23:13.888 This establishes a feedback loop for continuous application enhancements. 01:23:14.356 --> 01:23:19.461 The factory now has an end-to-end framework to adapt to changing conditions - often and easily. 01:23:20.362 --> 01:23:27.669 We make it easy to adapt and optimize NGC pretrained models with NVIDIA TAO and deploy and orchestrate applications with Fleet Command. 01:23:30.505 --> 01:23:36.945 We have all kinds of computer vision, speech, language and robotics models, and more coming all the time. 01:23:37.545 --> 01:23:39.948 Some are the work of our genetics and medical imaging team. 01:23:40.482 --> 01:23:45.754 For example, this model that predicts supplemental oxygen needs from xX-ray and electronic health records. 01:23:46.287 --> 01:23:52.727 This is a collaboration of 20 hospitals across 8 countries and 5 continents for COVID-19. 01:23:53.028 --> 01:23:56.898 Federated learning and NVIDIA TAO was essential to make this possible. 01:23:58.533 --> 01:24:04.072 The world needs a state-of-the-art conversational AI that can be customized and processed anywhere. 01:24:04.706 --> 01:24:16.373 Today, we're announcing the availability of NVIDIA Jarvis - a state-of-the-art deep learning AI for speech recognition, language understanding, translations, and speech. 01:24:17.252 --> 01:24:27.949 End-to-end GPU-accelerated, Jarvis interacts in about 100 milliseconds - listening, understanding, and responding faster than the blink of a human eye. 01:24:28.930 --> 01:24:34.718 We trained Jarvis for several million GPU-hours, on over 1 billion pages of text, 01:24:34.718 --> 01:24:41.293 and sixty thousand hours of speech in different languages, accents, environments and lingos. 01:24:41.810 --> 01:24:46.305 Out of the box, Jarvis achieves a world-class 90% recognition accuracy. 01:24:46.305 --> 01:24:51.568 You can get even better for your application by refining with your own data using NVIDIA TAO. 01:24:52.520 --> 01:24:57.792 Jarvis supports 5 languages today - English, Japanese, Spanish, German, French, and Russian. 01:24:58.560 --> 01:25:12.784 Jarvis does excellent translation, in the Bilingual Evaluation Understudy benchmark, Jarvis scores 40 points for English-to-Japanese, and 50 points for English-to-Spanish. 01:25:13.608 --> 01:25:18.780 40 is high quality and state-of-the-art. 50 is considered fluent translation. 01:25:19.881 --> 01:25:26.588 Jarvis can be customized for domain jargon. We've trained Jarvis for technical and healthcare scenarios. It's easy with TAO. 01:25:26.588 --> 01:25:33.394 And Jarvis now speaks with expression and emotion that you can control - no more mechanical talk. 01:25:33.862 --> 01:25:44.172 And lastly, this is a big one - Jarvis can be deployed in the cloud, on EGX in your data center, at the edge in a shop, warehouse, or 01:25:44.172 --> 01:25:49.944 factory running on EGX Aerial, or inside a delivery robot running on a Jetson computer. 01:25:50.712 --> 01:25:58.920 Jarvis early access program began last May and our Conversational AI software has been downloaded 45,000 times. 01:25:59.587 --> 01:26:08.796 Among early users is T-Mobile, the U.S. telecom giant. They're using Jarvis to offer exceptional customer service that demands high-quality 01:26:08.796 --> 01:26:12.000 and low-latency needed in real-time speech recognition. 01:26:12.967 --> 01:26:21.305 We're announcing a partnership with Mozilla Common Voice, one of the world's largest multi-language voice datasets, and its openly available to all 01:26:22.076 --> 01:26:33.348 NVIDIA will use our DGXs to process and train Jarvis with the dataset of 150,000 speakers in 65 languages and offer Jarvis back to the community for free. 01:26:33.655 --> 01:26:36.984 So go to Mozilla Common Voice and make some recordings! 01:26:36.984 --> 01:26:42.789 Let's make universal translation possible and help people around the world understand each other. 01:26:43.498 --> 01:26:44.866 Now let me show you Jarvis. 01:26:45.967 --> 01:26:48.069 The first part of Jarvis is speech recognition. 01:26:48.303 --> 01:26:51.870 Jarvis is over 90% accurate out-of-the-box - that's world-class. 01:26:51.870 --> 01:26:58.001 And you can still use TAO to make it even better for your application, like customizing for healthcare jargon. 01:26:58.846 --> 01:27:05.887 Chest x-ray shows left retrocardiac opacity, and this may be due to atelectasis, aspiration, or early pneumonia. 01:27:07.388 --> 01:27:10.592 I have no idea what I said but Jarvis recognized it perfectly. 01:27:11.226 --> 01:27:13.795 Jarvis translation now supports five languages. 01:27:13.995 --> 01:27:15.149 Let's do Japanese. 01:27:15.149 --> 01:27:25.990 "Excuse me, I'm looking for the famous Jangara Ramen shop. It should be nearby, but I don't see it on my map. Can you show me the way? I'm very hungry." 01:27:28.710 --> 01:27:33.615 That's great. Excellent accuracy. I think. Instantaneous response. 01:27:34.082 --> 01:27:37.285 You can do German, French, and Russian--with more languages on the way. 01:27:37.952 --> 01:27:39.554 Jarvis also speaks with feelings. 01:27:40.188 --> 01:27:44.025 Let's try this - "The more you buy, the more you save!" 01:27:48.596 --> 01:27:51.499 "The more you buy, the more you save" 01:27:52.033 --> 01:27:53.935 I think we are going to need more enthusiasm. 01:27:59.741 --> 01:28:02.710 "The more you buy, the more you save!" 01:28:02.910 --> 01:28:13.187 NVIDIA Jarvis - state-of-the-art deep learning conversational AI, interactive response, 5 languages, customize with TAO, 01:28:13.187 --> 01:28:16.697 and deploy from cloud, to edge, to autonomous systems. 01:28:17.492 --> 01:28:24.866 Recommender systems are the most important machine learning pipeline in the world today. 01:28:24.866 --> 01:28:29.707 It is the engine for search, ads, on-line shopping, music, books, movies, user-generated content, news. 01:28:30.538 --> 01:28:38.279 Recommender systems predict your needs and preferences from past interactions with you, your explicit preferences 01:28:38.279 --> 01:28:41.049 and learned preferences using methods called collaborative and content filtering. 01:28:42.283 --> 01:28:47.422 Trillions of items to be recommended to billions of people - the problem space is quite large. 01:28:48.256 --> 01:28:55.327 We would like to productize a state-of-the-art recommender system so that all companies can benefit from the transformative capabilities of this AI. 01:28:56.097 --> 01:29:06.603 We built an open-source recommender system framework, called Merlin, which simplifies an end-to-end workflow from ETL, to training, to validation, to inference. 01:29:07.375 --> 01:29:10.178 It is architected to scale as your dataset grows. 01:29:10.745 --> 01:29:15.450 And if you already have a ton of data, it is the fastest recommender system ever built. 01:29:16.050 --> 01:29:27.328 In our benchmarks, we achieved speed-ups of 10-50x for ETL, 2-10x for training and 3-100 times for inference depending on the exact setup. 01:29:28.663 --> 01:29:31.432 Merlin is now available on NGC. 01:29:32.033 --> 01:29:39.974 Our vision for Maxine is to eventually be your avatar in virtual worlds created in Omniverse - to enable virtual presence. 01:29:40.975 --> 01:29:44.011 The technologies of virtual presence can benefit video conferencing today. 01:29:44.545 --> 01:29:46.347 Alex is going to tell you all about it. 01:29:47.382 --> 01:29:51.219 Hi everyone, I'm Alex and I'm the Product Manager for NVIDIA Maxine. 01:29:51.819 --> 01:29:57.959 These days, we're spending so much time doing video conferencing. Don't you want a much better video communication experience? 01:29:58.760 --> 01:30:07.418 Today, I am so thrilled today to share with you the NVIDIA Maxine, a suite of AI technologies that can improve the video conferencing experience for everyone. 01:30:08.035 --> 01:30:13.741 When combined with NVIDIA Jarvis, Maxine offers the most accurate speech to text . 01:30:13.741 --> 01:30:17.612 See, It's now transcribing everything I'm saying, all in real time. 01:30:18.446 --> 01:30:25.653 And in addition, with the help of Jarvis, Maxine can also translate what I am saying into multiple languages. 01:30:26.254 --> 01:30:29.190 That would make international meetings so much easier. 01:30:30.024 --> 01:30:34.429 Another great feature I'd love to share with you is Maxine's Eye Contact feature. 01:30:35.363 --> 01:30:39.066 Often times when I am presenting, I am not looking at the camera 01:30:39.467 --> 01:30:47.241 When I turn on Maxine's eye contact feature, it corrects the position of my eyes so that I'm looking back into the camera again. 01:30:47.241 --> 01:30:49.444 Now, I can make eye contact with everyone else. 01:30:49.844 --> 01:30:52.814 The meeting experience just gets so much more engaging! 01:30:53.114 --> 01:30:59.987 Last but definitely not least, Maxine can even improve the video quality when bandwidth isn't sufficient. 01:31:00.321 --> 01:31:06.627 Now let's take a look at my video quality when bandwidth drops to as low as 50 kbps. 01:31:06.627 --> 01:31:08.162 Definitely not great. 01:31:08.596 --> 01:31:16.103 Maxine's AI face codec can improve the quality of my video even when bandwidth is as low as 50 kbps. 01:31:17.405 --> 01:31:26.914 The most powerful part of Maxine is that all of these amazing AI features can run simultaneously together and all in real time. 01:31:26.914 --> 01:31:30.384 Or you can just pick and choose a few. It's just that flexible! 01:31:30.885 --> 01:31:40.394 Now before I go let me turn off all the Maxine features so you can see what I looked like without Maxine's help. Not ideal. 01:31:40.394 --> 01:31:42.864 Now let's turn back on all the Maxine features. 01:31:43.698 --> 01:31:47.568 See - so much better thanks to NVIDIA Maxine! 01:31:49.904 --> 01:31:51.539 NGC has pre-trained models. 01:31:52.273 --> 01:31:55.676 TAO lets you fine-tune and adapt models to your applications. 01:31:56.043 --> 01:31:58.479 Fleet Command deploys and orchestrates your models. 01:31:58.880 --> 01:32:07.613 The final piece is the inference server - to infer insight from the continuous streams of data coming into your EGX servers or your cloud instance. 01:32:08.890 --> 01:32:10.958 NVIDIA Triton is our inference server. 01:32:11.459 --> 01:32:16.564 Triton is a model scheduling and dispatch engine that can handle just about anything you throw at it: 01:32:17.164 --> 01:32:21.469 Any AI model that runs on cuDNN, so basically every AI model. 01:32:22.103 --> 01:32:30.878 From any framework - TensorFlow, Pytorch, ONNX, OpenVINO, TensorRT, or custom C++/python backends. 01:32:31.178 --> 01:32:35.850 Triton schedules on the multiple generations of NVIDIA GPUs and x86 CPUs. 01:32:36.417 --> 01:32:39.921 Triton maximizes the CPU and GPU utilization. 01:32:40.655 --> 01:32:44.292 Triton scales with Kubernetes and handles live updates. 01:32:44.959 --> 01:32:51.499 Triton is fantastic for everything from image to speech recognition, from recommenders to language understanding. 01:32:52.099 --> 01:32:56.871 But let me show you something really hard - generating biomolecules for drug discovery. 01:32:57.505 --> 01:33:06.493 In drug discovery, it all comes down to finding the right molecule, the right shape, the right interactions with a protein - the right pharmaco-kinetic properties. 01:33:07.548 --> 01:33:16.000 Working with scientists from Astra-Zeneca, NVIDIA researchers trained a language model to understand SMILES - the language of chemical structures. 01:33:16.991 --> 01:33:20.661 We developed the model with Megatron and trained it on a SuperPOD. 01:33:20.861 --> 01:33:32.947 Then created a generative model that reads the structure of successful chemical drug-compounds to then generate potentially effective novel-compounds. 01:33:33.874 --> 01:33:45.824 Now we can use AI to generate candidate compounds that can then be further refined with physics-based simulations, like docking or Schrödinger's FEP+. 01:33:47.455 --> 01:33:52.994 Generating with a CPU is slow - it takes almost 8 seconds to generate one molecule. 01:33:53.961 --> 01:34:01.402 On a single A100 with Triton, it takes about 0.3 seconds - 32X faster! 01:34:02.603 --> 01:34:08.909 Using Triton Inference Server, we can scale this up to a SuperPOD and generate thousands of molecules per second. 01:34:09.877 --> 01:34:13.981 AI and simulation is going to revolutionize drug discovery. 01:34:14.649 --> 01:34:19.578 We provide these AI capabilities to our ecosystem of millions of developers around the world. 01:34:19.578 --> 01:34:22.954 Thousands of companies are building their most important services on NVIDIA AI. 01:34:23.691 --> 01:34:28.195 BestBuy is using Morpheus as the foundation of AI-based anomaly detection in their network. 01:34:29.063 --> 01:34:37.104 GE Healthcare has built an echocardiogram that uses TensorRT to quickly identify different views of the wall motion of the heart and select the best ones for analysis. 01:34:39.340 --> 01:34:46.405 Spotify has over 4 billion playlists. They're using RAPIDS to analyze models more efficiently, which lets them refresh personalized playlists more often. 01:34:46.410 --> 01:34:50.718 This keeps their content fresh with the latest trends. 01:34:51.519 --> 01:34:59.427 WeChat is using TensorRT to achieve 8msmillisecond latency even when using the state-of-the-art natural language understanding models like BERT. 01:34:59.427 --> 01:35:03.731 This gives them more accuracy and relevant results when maintaining real-time response. 01:35:04.765 --> 01:35:09.303 It's great to see NVIDIA AI used to build these amazing AI services. 01:35:10.204 --> 01:35:19.246 NVIDIA AI on EGX with Aerial 5G the AI computing platform for the enterprise and Edge - the next wave of AI. 01:35:20.881 --> 01:35:23.851 DRIVE AV is our AV platform and service. 01:35:24.685 --> 01:35:31.158 AV is the most intense machine learning and robotics challenge - one of the hardest, but also the greatest impact. 01:35:31.826 --> 01:35:40.735 We're building an AV platform end-to-end - from the AV chips and computers, sensor architecture, data processing, mapping, developing the 01:35:40.735 --> 01:35:46.914 driving software, creating the simulator and digital twin, fleet command operations, to road testing. 01:35:46.914 --> 01:35:51.445 All of it to the highest functional safety and cybersecurity standards. safety and cybersecurity standards. 01:35:51.946 --> 01:35:54.448 None of these things are openly available, so we build them. 01:35:55.249 --> 01:36:03.457 We build in the modules so our customers and partners in the $10 trillion transportation industry can leverage the parts they need. 01:36:04.024 --> 01:36:08.195 Meanwhile, we build our end-to-end AV service in partnership with Mercedes. 01:36:08.696 --> 01:36:10.798 AV computing demand is skyrocketing. 01:36:10.831 --> 01:36:13.334 Orin is a giant leap over Xavier. 01:36:13.934 --> 01:36:20.307 The more developers learn about AV, the more advanced the algorithms become, and the greater the computing demand. 01:36:21.242 --> 01:36:28.649 More computation capacity gives teams faster iteration and speed to market - leading some to call TOPS the new horsepower. 01:36:29.083 --> 01:36:35.422 And many are realizing that the reserved computing capacity today is an opportunity to offer new services tomorrow. 01:36:36.056 --> 01:36:41.362 NVIDIA DRIVE is an open programmable platform that AI engineers all over the world are familiar with. 01:36:41.896 --> 01:36:48.536 New AI technology is being invented on NVIDIA AI constantly - these inventions will be tomorrow's new services. 01:36:49.470 --> 01:36:52.306 Orin is an amazing AV computer. 01:36:52.306 --> 01:36:57.812 Today, we are announcing that Orin was also designed to be the central computer of the car. 01:36:58.312 --> 01:37:08.122 Orin will process in one central computer the cluster, infotainment, passenger interaction AI, and very importantly, the confidence view 01:37:08.289 --> 01:37:10.057 or the perception world model. 01:37:10.424 --> 01:37:21.061 The confidence view is what the car actually perceives around it and construct it into a 3D surround model - this is what is in the mind of the autopilot AI. 01:37:21.101 --> 01:37:26.941 It is important that the car shows us that its surround perception is accurate, so that we have confidence in its driving. 01:37:27.675 --> 01:37:31.378 In time, rear view mirrors will be replaced by the surround 3D perception. 01:37:32.213 --> 01:37:42.356 The future is one central computer - 4 domains, virtualized and isolated, architected for functional safety and security, software-defined 01:37:42.623 --> 01:37:48.662 and upgradeable for the life of the car - and super smart AI and beautiful graphics. 01:37:49.897 --> 01:37:57.304 Many car companies have already adopted Orin and so it would help them tremendously to also leverage our full AV development platform. 01:37:58.105 --> 01:38:08.382 Today, we are announcing NVIDIA’s 8th generation Hyperion car platform - including reference sensors, AV and central computers, the 3D ground 01:38:08.415 --> 01:38:12.119 truth data recorder, networking, and all of the essential software. 01:38:13.153 --> 01:38:19.260 Hyperion 8 is compatible with the NVIDIA DRIVE AV stack, so easy to adopt and integrate elements of our stack. 01:38:19.793 --> 01:38:28.435 Hyperion 8 is a fully functional, going-to-production, open AV platform for the multi-trillion-dollar transportation ecosystem. 01:38:29.470 --> 01:38:32.206 Orin will be in production in 2022. 01:38:32.473 --> 01:38:38.479 Meanwhile, our next generation is already in full gear and will be yet another giant leap. 01:38:39.179 --> 01:38:42.549 Today, we are announcing NVIDIA DRIVE Atlan. 01:38:43.050 --> 01:38:52.660 DRIVE Atlan will be 1,000 TOPS on one chip - more than the total compute in most Level 5 robotaxis today. 01:38:53.394 --> 01:39:02.603 To achieve higher autonomy in more conditions, sensor resolutions will continue to increase. There will be more of them. AI models will get 01:39:02.603 --> 01:39:09.543 more sophisticated. There will be more redundancy and safety functionality. We're going to need all of the computing we can get. 01:39:10.711 --> 01:39:21.023 Atlan is a technical marvel - fusing all of NVIDIA's technologies in AI, auto, robotics, safety, and Bluefield secure data center technologies. 01:39:21.755 --> 01:39:33.666 AV and software must be planned as multi-generational investments. The software you invested billions in today must carry over to the entire fleet and to future generations. 01:39:34.201 --> 01:39:38.906 The DRIVE Atlan - The Next Level - One Architecture. 01:39:40.541 --> 01:39:42.876 The car industry has become a technology industry. 01:39:43.344 --> 01:39:48.315 Future cars are going to be completely programmable computers and business models are going to be software driven. 01:39:48.916 --> 01:39:52.252 Car companies will offer software services for the life of the car. 01:39:53.053 --> 01:40:02.463 The new tech mindset sees the car not just as a product to sell, but as an installed base of 10's or 100's of millions to build upon, 01:40:03.063 --> 01:40:06.100 creating billions in services and opportunities. 01:40:07.267 --> 01:40:09.670 The world's big brands have giant opportunities. 01:40:10.337 --> 01:40:12.106 Step one was going electric. 01:40:12.573 --> 01:40:16.110 Now the big brands are making their new fleets autonomous and programmable. 01:40:16.110 --> 01:40:19.613 Orin will be powering many next generation EVs. 01:40:20.247 --> 01:40:23.017 And with Mercedes, we are building the end-to-end system. 01:40:24.184 --> 01:40:26.587 The world moves 10 trillion miles a year. 01:40:27.588 --> 01:40:32.092 If only a fraction of these miles were served by robotaxis, the opportunity is giant. 01:40:33.227 --> 01:40:36.030 We expect robotaxis to start ramping in the next couple years. 01:40:36.864 --> 01:40:43.670 These services will also be platforms for all kinds of new services to be built upon - like last mile delivery. 01:40:45.005 --> 01:40:49.276 The Internet moves electrons, trucks move atoms. 01:40:50.044 --> 01:40:59.219 The rise of e-commerce is putting intense pressure on this system - one click and a new TV shows up at your house, another click and a 01:40:59.219 --> 01:41:00.287 cheeseburger shows up. 01:41:01.055 --> 01:41:02.523 This trend will only go up. 01:41:03.257 --> 01:41:05.993 The U.S. will be short by 100K truckers by 2023. 01:41:07.194 --> 01:41:09.997 The EU is short 150K truckers already. 01:41:10.831 --> 01:41:12.733 China is short 4 million drivers. 01:41:13.500 --> 01:41:24.054 So ideas like driverless trucks from hub-to-hub or driverless trucks inside a port or warehouse campus are excellent near-term ways to augment with automation. 01:41:25.546 --> 01:41:28.415 We started my talk with Omniverse, we'll close with Omniverse. 01:41:29.683 --> 01:41:38.632 You can see how important it is to our work, to robotics, to anyone building AIs that interact with the physical world, to have a physically based simulator, or digital twin. 01:41:39.560 --> 01:41:43.363 In the case of Isaac, the digital twin is the factory in Omniverse. 01:41:44.198 --> 01:41:48.869 In the case of DRIVE, the digital twin is the collective memories of the fleet captured in Omniverse. 01:41:49.670 --> 01:41:57.077 The DRIVE Digital Twin is used throughout the development - it's used for HD map reconstruction, synthetic data generation so we can 01:41:57.077 --> 01:42:05.986 bootstrap training new models, new scenario simulations, hardware-in-the-loop simulations, release validation, for replaying unfamiliar 01:42:06.019 --> 01:42:12.426 scenarios experienced by a car, or a teleoperator up-linking into a car to remotely pilot. 01:42:13.360 --> 01:42:18.599 The DRIVE Digital Twin in Omniverse is a virtual world that every car in the fleet is connected to. 01:42:19.333 --> 01:42:26.440 Today, we're announcing that DRIVE Sim, the engine of DRIVE Digital Twin, will be available for the community this summer. 01:42:27.107 --> 01:42:31.612 Let me show you what DRIVE AV and DRIVE Sim can do - this is a mountain of technology. 01:42:31.879 --> 01:42:32.379 Enjoy. 01:45:06.033 --> 01:45:11.305 What a GTC! It's incredible to think how much was done in this last year. 01:45:11.605 --> 01:45:13.173 I spoke about several things: 01:45:13.774 --> 01:45:16.443 NVIDIA is a full-stack computing platform. 01:45:17.077 --> 01:45:25.845 We're building virtual worlds with NVIDIA Omniverse - a miraculous platform that will help build the next wave of AI for robotics and self-driving cars, 01:45:27.087 --> 01:45:38.255 We announced new DGX systems and new software. Megatron for giant Transformers, Clara for drug discovery, and cuQuantum for quantum computing. 01:45:39.466 --> 01:45:51.299 NVIDIA is now a 3-chip company. With the addition of Grace CPU, designed for a specialized segment of computing focused on giant-scale AI and HPC. 01:45:52.212 --> 01:45:58.185 And for data center infrastructure processing, we announced BlueField-3 and DOCA 1.0. 01:45:58.885 --> 01:46:01.121 NVIDIA is expanding the reach of AI. 01:46:01.888 --> 01:46:11.465 We announced an important new platform, NVIDIA EGX with Aerial 5G, to make AI accessible to all companies and industries. 01:46:12.466 --> 01:46:21.274 We're joined by the leaders of IT - VMware, computer makers, infrastructure software, storage, and security providers. 01:46:21.975 --> 01:46:28.281 And to lower the bar to the highest levels of AI, we offer Pre-Trained Models - like 01:46:28.281 --> 01:46:29.916 Jarvis conversational AI, 01:46:30.417 --> 01:46:31.918 Merlin recommender system, 01:46:32.819 --> 01:46:34.554 Maxine virtual conferencing, 01:46:35.122 --> 01:46:36.656 and Morpheus AI security. 01:46:38.225 --> 01:46:41.361 All of it is optimized on NVIDIA AI Enterprise. 01:46:42.062 --> 01:46:48.935 And with new tools like NVIDIA TAO, Fleet Command, and Triton, it's easy for you to customize and deploy on EGX. 01:46:49.803 --> 01:46:59.713 And Drive - our end-to-end platform for the $10 trillion dollar transportation industry, which is becoming one of the world's largest technology industries. technology industries. 01:47:00.680 --> 01:47:14.861 From Orin and Atlan, the first 1000 TOPS SOC, to Hyperion 8, a fully-operational reference AV-car platform, to Drive Sim, to Drive AV - 01:47:14.861 --> 01:47:16.997 we're working with the industry at every layer. 01:47:18.198 --> 01:47:26.907 20 years ago, all of this was science fiction. 10 years ago, it was a dream. Today, we're living it. 01:47:28.241 --> 01:47:33.847 I want to thank all of you - developers and partners, and a very special thanks to all the NVIDIA employees. 01:47:34.815 --> 01:47:42.255 Our purpose is to advance our craft, so that you, the Da Vincis of our time, can advance yours. 01:47:43.089 --> 01:47:48.895 Ultimately, NVIDIA is an instrument, an instrument for you to do your life's work. 01:47:50.163 --> 01:47:51.364 Have a great GTC.