For nearly four decades\u00a0Moore\u00a0's Law has been the governing dynamics\u00a0of the computer industry\u00a0which in turn has impacted every industry.\u00a0The exponential performance increase\u00a0at constant cost and power has slowed.\u00a0Yet, computing advance has gone to lightspeed.\u00a0The warp drive engine is accelerated computing\u00a0and the energy source is AI.\u00a0The arrival of accelerated computing and AI is timely\u00a0as industries tackle powerful dynamics\u00a0sustainability\u00a0generative AI\u00a0and digitalization.\u00a0Without Moore\u00a0's Law, as computing surges,\u00a0data center power is skyrocketing\u00a0and companies struggle to achieve Net Zero.\u00a0The impressive capabilities of Generative AI\u00a0created a sense of urgency for companies to reimagine their products and business models.\u00a0Industrial companies are racing to digitalize and reinvent into software-driven tech companies\u00a0to be the disruptor\u00a0and not the disrupted.\u00a0Today, we will discuss how accelerated computing and AI are powerful tools for tackling these challenges\u00a0and engaging the enormous opportunities ahead.\u00a0We will share new advances in NVIDIA\u00a0's full-stack, datacenter-scale, accelerated computing platform.\u00a0We will reveal new chips and systems,\u00a0acceleration libraries, cloud and AI services\u00a0and partnerships that open new markets.\u00a0Welcome to GTC!\u00a0GTC is our conference for developers.\u00a0The global NVIDIA ecosystem spans 4 million developers,\u00a040,000 companies\u00a0and 14,000 startups.\u00a0Thank you to our Diamond sponsors for supporting us\u00a0and making GTC 2023 a huge success.\u00a0We\u00a0're so excited to welcome more than 250,000\u00a0of you to our conference.\u00a0GTC has grown incredibly.\u00a0Only four years ago, our in-person GTC\u00a0conference had 8,000 attendees.\u00a0At GTC 2023, we\u00a0'll learn from leaders\u00a0like Demis Hassabis of DeepMind\u00a0Valeri Taylor of Argonne Labs\u00a0Scott Belsky of Adobe\u00a0Paul Debevec of Netflix\u00a0Thomas Schulthess of ETH Zurich\u00a0and a special fireside chat I\u00a0'm having with Ilya Sutskever\u00a0co-founder of OpenAI, the creator of ChatGPT.\u00a0We have 650 amazing talks from the brightest minds in academia and the world\u00a0's largest industries:\u00a0There are more than 70 talks on Generative AI alone.\u00a0Other great talks, like pre-trained multi-task models for robotics\u2026\u00a0sessions on synthetic data generation, an important method for advancing AI\u00a0including one on using Isaac Sim to generate physically\u00a0based lidar point clouds\u00a0a bunch of talks on digital twins, from using AI\u00a0to populate virtual factories of the future\u00a0to restoring lost Roman mosaics of the past\u00a0cool talks on computational instruments, including a giant optical telescope and a photon-counting CT\u00a0materials science for carbon capture and solar cells,\u00a0to climate science, including our work on Earth-2\u00a0important works by NVIDIA Research\u00a0on trustworthy AI and AV safety\u00a0From computational lithography for micro-chips,\u00a0to make the smallest machines\u00a0to AI at the Large Hadron Collider to explain the universe.\u00a0The world\u00a0's most important companies are here\u00a0from auto and transportation\u00a0healthcare, manufacturing, financial services,\u00a0retail, apparel, media and entertainment, telco\u00a0and of course, the world\u00a0's leading AI companies.\u00a0The purpose of GTC is to inspire the world on the art-of-the-possible of accelerating computing\u00a0and to celebrate the achievements of the scientists\u00a0and researchers that use it.\u00a0I am a translator.\u00a0Transforming text into creative discovery,\u00a0movement into animation,\u00a0and direction into action.\u00a0I am a healer.\u00a0Exploring the building blocks that make us unique\u00a0modeling new threats before they happen\u00a0and searching for the cures to keep them at bay.\u00a0I am a visionary.\u00a0Generating new medical miracles\u00a0and giving us a new perspective on our sun\u00a0to keep us safe here on earth.\u00a0I am a navigator.\u00a0Discovering a unique moment in a sea of content\u00a0we\u00a0're announcing the next generation\u00a0and the perfect setting for any story.\u00a0I am a creator.\u00a0Building 3D experiences from snapshots\u00a0and adding new levels of reality to our virtual selves.\u00a0I am a helper.\u00a0Bringing brainstorms to life\u00a0sharing the wisdom of a million programmers\u00a0and turning ideas into virtual worlds.\u00a0Build northern forest.\u00a0I even helped write this script\u00a0breathed life into the words\u00a0and composed the melody.\u00a0I am AI.\u00a0Brought to life by NVIDIA, deep learning,\u00a0and brilliant minds everywhere.\u00a0NVIDIA invented accelerated computing to solve problems\u00a0that normal computers can\u00a0't.\u00a0Accelerated computing is not easy\u00a0it requires full-stack invention from chips, systems, networking,\u00a0acceleration libraries, to refactoring the applications.\u00a0Each optimized stack accelerates an application domain\u00a0from graphics, imaging, particle or fluid dynamics\u00a0quantum physics, to data processing and machine learning.\u00a0Once accelerated, the application can enjoy incredible speed-up, as well as scale-up across many computers.\u00a0The combination of speed-up and scale-up\u00a0has enabled us to achieve a million-X\u00a0for many applications over the past decade\u00a0helping solve problems previously impossible.\u00a0Though there are many examples,\u00a0the most famous is deep learning.\u00a0In 2012, Alex Kerchevsky, Ilya Suskever, and Geoff Hinton\u00a0needed an insanely fast computer\u00a0to train the AlexNet computer vision model.\u00a0The researchers trained AlexNet with\u00a014 million images on GeForce GTX 580\u00a0\u00a0processing 262 quadrillion floating-point operations,\u00a0and the trained model won the ImageNet challenge by a wide margin, and ignited the Big Bang of AI.\u00a0A decade later, the transformer model was invented.\u00a0And Ilya, now at OpenAI, trained the GPT-3\u00a0large language model to predict the next word.\u00a0323 sextillion floating-point operations were required to train GPT-3.\u00a0One million times more floating-point operations\u00a0than to train AlexNet.\u00a0The result this time \u2013 ChatGPT, the AI heard around the world.\u00a0A new computing platform has been invented.\u00a0The iPhone moment of AI has started.\u00a0Accelerated computing and AI have arrived.\u00a0Acceleration libraries are at the core of accelerated computing.\u00a0These libraries connect to applications which connect to the world\u00a0's industries, forming a network of networks.\u00a0Three decades in the making, several thousand applications\u00a0are now NVIDIA accelerated\u00a0with libraries in almost every domain of science and industry.\u00a0All NVIDIA GPUs are CUDA-compatible, providing a large install base and significant reach for developers.\u00a0A wealth of accelerated applications attract end users, which creates a large market for cloud service providers\u00a0and computer makers to serve.\u00a0A large market affords billions in R&D to fuel its growth.\u00a0NVIDIA has established the accelerated computing virtuous cycle.\u00a0Of the 300 acceleration libraries and 400 AI models\u00a0that span ray tracing and neural rendering\u00a0physical, earth, and life sciences, quantum physics\u00a0and chemistry, computer vision\u00a0data processing, machine learning and AI, we updated 100\u00a0we updated 100 this year that increase performance\u00a0and features for our entire installed base.\u00a0Let me highlight some acceleration libraries that solve new challenges and open new markets.\u00a0The auto and aerospace industries use CFD for turbulence\u00a0and aerodynamics simulation.\u00a0The electronics industry uses CFD for thermal management design.\u00a0This is Cadence\u00a0's slide of their new CFD solver\u00a0accelerated by CUDA.\u00a0At equivalent system cost, NVIDIA A100 is 9X\u00a0the throughput of CPU servers.\u00a0Or at equivalent simulation throughput, NVIDIA is 9X lower cost\u00a0or 17X less energy consumed.\u00a0Ansys, Siemens, Cadence, and other leading CFD solvers\u00a0are now CUDA-accelerated.\u00a0Worldwide, industrial CAE uses nearly\u00a0100 billion CPU core hours yearly.\u00a0Acceleration is the best way to reclaim power and achieve sustainability and Net Zero.\u00a0NVIDIA is partnering with the global quantum\u00a0computing research community.\u00a0The NVIDIA Quantum platform consists of libraries and systems for researchers to advance quantum programming models,\u00a0system architectures, and algorithms.\u00a0cuQuantum is an acceleration library\u00a0for quantum circuit simulations.\u00a0IBM Qiskit, Google Cirq, Baidu Quantum Leaf, QMWare, QuEra, Xanadu Pennylane, Agnostiq, and AWS Bracket\u00a0have integrated cuQuantum into their simulation frameworks.\u00a0Open Quantum CUDA is our hybrid GPU-Quantum\u00a0programming model.\u00a0IonQ, ORCA Computing, Atom, QuEra, Oxford Quantum Circuits, IQM, Pasqal, Quantum Brilliance, Quantinuum, Rigetti,\u00a0Xanadu, and Anyon have integrated Open Quantum CUDA.\u00a0Error correction on a large number of qubits is necessary to recover data from quantum noise and decoherence.\u00a0Today, we are announcing a quantum control link, developed in partnership with Quantum Machines\u00a0that connects NVIDIA GPUs to a quantum computer to\u00a0do error correction at extremely high speeds.\u00a0Though commercial quantum computers are still a decade or two away, we are delighted to support this large and vibrant\u00a0research community with NVIDIA Quantum.\u00a0Enterprises worldwide use Apache Spark to\u00a0process data lakes and warehouses\u00a0SQL queries, graph analytics, and recommender systems.\u00a0Spark-RAPIDS is NVIDIA\u00a0's accelerated Apache Spark\u00a0data processing engine.\u00a0Data processing is the leading workload\u00a0of the world\u00a0's $500B cloud computing spend.\u00a0Spark-RAPIDS now accelerates major cloud data processing platforms, including GCP Dataproc\u00a0Amazon EMR, Databricks, and Cloudera.\u00a0Recommender systems use vector databases to store, index, search, and retrieve massive datasets of unstructured data.\u00a0A new important use-case of vector databases is large language models to retrieve domain-specific or proprietary facts\u00a0that can be queried during text generation.\u00a0We are introducing a new library, RAFT,\u00a0to accelerate indexing, loading the data\u00a0and retrieving a batch of neighbors for a single query.\u00a0We are bringing the acceleration of RAFT to Meta\u00a0's open-source FAISS AI Similarity Search, Milvus open-source vector DB\u00a0used by over 1,000 organizations,\u00a0and Redis with over 4B docker pulls.\u00a0Vector databases will be essential for organizations building proprietary large language models.\u00a0Twenty-two years ago, operations research scientists Li and Lim posted a series of challenging pickup and delivery problems.\u00a0PDP shows up in manufacturing, transportation,\u00a0retail and logistics, and even disaster relief.\u00a0PDP is a generalization of the Traveling Salesperson Problem\u00a0and is NP-hard\u00a0meaning there is no efficient algorithm to find an exact solution.\u00a0The solution time grows factorially as the problem size increases.\u00a0Using an evolution algorithm and accelerated computing\u00a0to analyze 30 billion moves per second\u00a0\u00a0NVIDIA cuOpt has broken the world record and discovered\u00a0the best solution for Li&Lim\u00a0's challenge.\u00a0AT&T routinely dispatches 30,000 technicians to service 13 million customers across 700 geographic zones.\u00a0Today, running on CPUs, AT&T\u00a0's dispatch\u00a0optimization takes overnight.\u00a0AT&T wants to find a dispatch solution in real time that continuously optimizes for urgent customer needs\u00a0and overall customer satisfaction, while adjusting\u00a0for delays and new incidents that arise.\u00a0With cuOpt, AT&T can find a solution 100X faster\u00a0and update their dispatch in real time.\u00a0AT&T has adopted a full suite of NVIDIA AI libraries.\u00a0In addition to Spark-RAPIDS and cuOPT, they\u00a0're using Riva for conversational AI and Omniverse for digital avatars.\u00a0AT&T is tapping into NVIDIA accelerated computing and AI\u00a0for sustainability, cost savings, and new services.\u00a0cuOpt can also optimize logistic services. 400 billion parcels\u00a0are delivered to 377 billion stops each year.\u00a0Deloitte, Capgemini, Softserve, Accenture, and Quantiphi are using NVIDIA cuOpt to help customers optimize operations.\u00a0NVIDIA\u00a0's inference platform consists of three software SDKs.\u00a0NVIDIA TensorRT is our inference runtime\u00a0that optimizes for the target GPU.\u00a0NVIDIA Triton is a multi-framework data center inference serving software supporting GPUs and CPUs.\u00a0Microsoft Office and Teams, Amazon, American Express,\u00a0and the U.S. Postal Service\u00a0are among the 40,000 customers using TensorRT and Triton.\u00a0Uber uses Triton to serve hundreds of thousands\u00a0of ETA predictions per second.\u00a0With over 60 million daily users, Roblox uses Triton to serve models for game recommendations\u00a0build avatars, and moderate content and marketplace ads.\u00a0We are releasing some great new features \u2013 model analyzer support for model ensembles, multiple concurrent model serving,\u00a0and multi-GPU, multi-node inference\u00a0for GPT-3 large language models.\u00a0NVIDIA Triton Management Service is our new software that automates the scaling and orchestration\u00a0\u00a0of Triton inference instances across a data center.\u00a0Triton Management Service will help you improve the throughput and cost efficiency of deploying your models.\u00a050-80% of cloud video pipelines are processed on CPUs\u00a0consuming power and cost and adding latency.\u00a0CV-CUDA for computer vision, and VPF for video processing, are new cloud-scale acceleration libraries.\u00a0CV-CUDA includes 30 computer vision operators for detection, segmentation, and classification.\u00a0VPF is a python video encode and decode acceleration library.\u00a0Tencent uses CV-CUDA and VPF\u00a0to process 300,000 videos per day.\u00a0Microsoft uses CV-CUDA and VPF to process visual search.\u00a0Runway is a super cool company that uses\u00a0CV-CUDA and VPF to process video\u00a0for their cloud Generative AI video editing service.\u00a0Already, 80% of internet traffic is video.\u00a0User-generated video content is driving significant growth and consuming massive amounts of power.\u00a0We should accelerate all video processing and reclaim the power.\u00a0CV-CUDA and VPF are in early access.\u00a0NVIDIA accelerated computing helped\u00a0achieve a genomics milestone\u00a0now doctors can draw blood and sequence\u00a0a patient\u00a0's DNA in the same visit.\u00a0In another milestone, NVIDIA-powered instruments reduced the cost of whole genome sequencing to just $100.\u00a0Genomics is a critical tool in synthetic biology\u00a0with applications ranging from drug discovery\u00a0and agriculture to energy production.\u00a0NVIDIA Parabricks is a suite of AI-accelerated libraries for end-to-end genomics analysis in the cloud or in-instrument.\u00a0NVIDIA Parabricks is available in every public cloud and genomics platforms like Terra, DNAnexus, and FormBio.\u00a0Today, we\u00a0're announcing Parabricks 4.1 and will run on NVIDIA-accelerated genomics instruments\u00a0from PacBio, Oxford Nanopore, Ultima,\u00a0Singular, BioNano, and Nanostring.\u00a0The world\u00a0's $250B medical instruments\u00a0market is being transformed.\u00a0Medical instruments will be software-defined and AI powered.\u00a0NVIDIA Holoscan is a software library\u00a0for real-time sensor processing systems.\u00a0Over 75 companies are developing\u00a0medical instruments on Holoscan.\u00a0Today, we are announcing Medtronic, the world leader in medical instruments, and NVIDIA are building their AI platform\u00a0for software-defined medical devices.\u00a0This partnership will create a common platform for Medtronic systems, ranging from surgical navigation\u00a0to robotic-assisted surgery.\u00a0Today, Medtronic announced that its next-generation GI Genius system, with AI for early detection of colon cancer\u00a0is built on NVIDIA Holoscan and\u00a0will ship around the end of this year.\u00a0The chip industry is the foundation of nearly every industry.\u00a0Chip manufacturing demands extreme precision, producing features 1,000 times smaller than a bacterium\u00a0and on the order of a single gold atom or a strand of human DNA.\u00a0Lithography, the process of creating patterns on a wafer, is the beginning of the chip manufacturing process\u00a0and consists of two stages \u2013 photomask making\u00a0and pattern projection.\u00a0It is fundamentally an imaging problem at the limits of physics.\u00a0The photomask is like a stencil of a chip. Light is blocked or passed through the mask\u00a0to the wafer to create the pattern.\u00a0The light is produced by the ASML EUV\u00a0extreme ultraviolet lithography system.\u00a0Each system is more than a quarter-of-a-billion dollars.\u00a0ASML EUV uses a radical way to create light.\u00a0Laser pulses firing 50,000 times a second at a drop of tin, vaporizing it, creating a plasma that emits 13.5nm EUV light\u00a0nearly X-ray.\u00a0Multilayer mirrors guide the light to the mask.\u00a0The multilayer reflectors in the mask reticle take advantage of interference patterns of the 13.5nm light\u00a0to create finer features down to 3nm.\u00a0Magic.\u00a0The wafer is positioned within a quarter of a nanometer and aligned 20,000 times a second to adjust for any vibration.\u00a0The step before lithography is equally miraculous.\u00a0Computational lithography applies inverse physics algorithms\u00a0to predict the patterns on the mask\u00a0\u00a0that will produce the final patterns on the wafer.\u00a0In fact, the patterns on the mask\u00a0do not resemble the final features at all.\u00a0Computational lithography simulates Maxwell\u00a0's equations\u00a0of the behavior of light passing through optics\u00a0and interacting with photoresists.\u00a0Computational lithography is the largest computation\u00a0workload in chip design and manufacturing\u00a0consuming tens of billions of CPU hours annually.\u00a0Massive data centers run 24\/7 to create reticles\u00a0used in lithography systems.\u00a0These data centers are part of the nearly $200 billion annual CAPEX invested by chip manufacturers.\u00a0Computational lithography is growing fast\u00a0as algorithm complexity increases\u00a0enabling the industry to go to 2nm and beyond.\u00a0NVIDIA today is announcing cuLitho, a library\u00a0for computational lithography.\u00a0cuLitho, a massive body of work that has taken nearly four years, and with close collaborations with TSMC,\u00a0ASML, and Synopsys, accelerates computational\u00a0lithography by over 40X.\u00a0There are 89 reticles for the NVIDIA H100.\u00a0Running on CPUs, a single reticle currently\u00a0takes two weeks to process.\u00a0cuLitho, running on GPUs, can process\u00a0a reticle in a single 8-hour shift.\u00a0TSMC can reduce their 40,000 CPU servers used for computational lithography by accelerating with cuLitho\u00a0on just 500 DGX H100 systems, reducing power\u00a0from 35MW to just 5MW.\u00a0With cuLitho, TSMC can reduce prototype cycle time,\u00a0increase throughput\u00a0and reduce the carbon footprint of their manufacturing,\u00a0and prepare for 2nm and beyond.\u00a0TSMC will be qualifying cuLitho for production starting in June.\u00a0Every industry needs to accelerate every workload, so that we can reclaim power and do more with less.\u00a0Over the past ten years, cloud computing has grown 20% annually into a massive $1T industry.\u00a0Some 30 million CPU servers do the majority of the processing.\u00a0There are challenges on the horizon.\u00a0As Moore\u00a0's Law ends, increasing CPU performance comes with increased power.\u00a0And the mandate to decrease carbon emissions is fundamentally at odds with the need to increase data centers.\u00a0Cloud computing growth is power-limited.\u00a0First and foremost, data centers must accelerate every workload.\u00a0Acceleration will reclaim power.\u00a0The energy saved can fuel new growth.\u00a0Whatever is not accelerated will be processed on CPUs.\u00a0The CPU design point for accelerated cloud datacenters\u00a0differs fundamentally from the past.\u00a0In AI and cloud services, accelerated computing offloads parallelizable workloads, and CPUs process other workloads,\u00a0like web RPC and database queries.\u00a0We designed the Grace CPU for an AI and cloud-first world,\u00a0where AI workloads are GPU-accelerated\u00a0and Grace excels at single-threaded execution\u00a0and memory processing.\u00a0It\u00a0's not just about the CPU chip. Datacenter operators optimize for throughput and total cost of ownership of the entire datacenter.\u00a0We designed Grace for high energy-efficiency\u00a0at cloud datacenter scale.\u00a0Grace comprises 72 Arm cores connected by a super high-speed on-chip scalable coherent fabric that delivers 3.2 TB\/sec\u00a0of cross-sectional bandwidth.\u00a0Grace Superchip connects 144 cores between two CPU dies over a 900 GB\/sec low-power chip-to-chip coherent interface.\u00a0The memory system is LPDDR low-power memory, like used in cellphones, that we specially enhanced for use in datacenters.\u00a0It delivers 1 TB\/s, 2.5x the bandwidth of today\u00a0's systems\u00a0at 1\/8th the power.\u00a0The entire 144-core Grace Superchip module\u00a0with 1TB of memory is only 5x8 inches.\u00a0It is so low power it can be air cooled.\u00a0This is the computing module with passive cooling.\u00a0Two Grace Superchip computers can fit\u00a0in a single 1U air-cooled server.\u00a0Grace\u00a0's performance and power efficiency are excellent for cloud and scientific computing applications.\u00a0We tested Grace on a popular Google benchmark, which tests how quickly cloud microservices communicate\u00a0and the Hi-Bench suite that tests Apache Spark\u00a0memory-intensive data processing.\u00a0These kinds of workloads are foundational for cloud datacenters.\u00a0At microservices, Grace is 1.3X faster than the average\u00a0of the newest generation x86 CPUs\u00a0and 1.2X faster at data processing\u00a0And that higher performance is achieved using only 60% of the power measured at the full server node.\u00a0CSPs can outfit a power-limited data center with 1.7X more Grace servers, each delivering 25% higher throughput.\u00a0At iso-power, Grace gives CSPs 2X the growth opportunity.\u00a0Grace is sampling.\u00a0And Asus, Atos, Gigabyte, HPE, QCT,\u00a0Supermicro, Wistron, and ZT are building systems now.\u00a0In a modern software-defined data center, the operating system doing virtualization, network, storage, and security can\u00a0consume nearly half of the datacenter\u00a0's CPU cores\u00a0and associated power.\u00a0Datacenters must accelerate every workload to reclaim power and free CPUs for revenue-generating workloads.\u00a0NVIDIA BlueField offloads and accelerates the datacenter operating system and infrastructure software.\u00a0Over two dozen ecosystem partners, including Check Point,\u00a0Cisco, DDN, Dell EMC\u00a0Juniper, Palo Alto Networks, Red Hat, and VMWare,\u00a0\u00a0use BlueField\u00a0's datacenter acceleration technology to run their software platforms more efficiently.\u00a0BlueField-3 is in production and adopted by leading cloud service providers, Baidu, CoreWeave, JD.com, Microsoft Azure,\u00a0Oracle OCI, and Tencent Games, to accelerate their clouds.\u00a0NVIDIA accelerated computing starts with DGX\u00a0the world\u00a0's AI supercomputer\u00a0the engine behind the large language model breakthrough.\u00a0I hand-delivered the world\u00a0's first DGX to OpenAI.\u00a0Since then, half of the Fortune 100 companies\u00a0have installed DGX AI supercomputers.\u00a0DGX has become the essential instrument of AI.\u00a0The GPU of DGX is eight H100 modules.\u00a0H100 has a Transformer Engine designed to process models\u00a0like the amazing ChatGPT,\u00a0which stands for Generative Pre-trained Transformers.\u00a0The eight H100 modules are NVLINK\u00a0'd to each other across NVLINK switches to allow fully non-blocking transactions.\u00a0The eight H100s work as one giant GPU.\u00a0The computing fabric is one of the most vital systems\u00a0of the AI supercomputer.\u00a0400 Gbps ultra-low latency NVIDIA Quantum InfiniBand\u00a0with in-network processing\u00a0connects hundreds and thousands of DGX nodes\u00a0into an AI supercomputer.\u00a0NVIDIA DGX H100 is the blueprint for\u00a0customers building AI infrastructure worldwide.\u00a0It is now in full production.\u00a0I am thrilled that Microsoft announced Azure is opening private previews to their H100 AI supercomputer.\u00a0Other systems and cloud services will soon come from Atos, AWS, Cirrascale, CoreWeave, Dell, Gigabyte, Google, HPE,\u00a0Lambda Labs, Lenovo, Oracle, Quanta, and SuperMicro.\u00a0The market for DGX AI supercomputers has grown significantly.\u00a0Originally used as an AI research instrument, DGX AI supercomputers are expanding into operation\u00a0running 24\/7 to refine data and process AI.\u00a0DGX supercomputers are modern AI factories.\u00a0We are at the iPhone moment of AI.\u00a0Start-ups are racing to build disruptive products and business models, while incumbents are looking to respond.\u00a0Generative AI has triggered a sense of urgency in enterprises worldwide to develop AI strategies.\u00a0Customers need to access NVIDIA AI easier and faster.\u00a0We are announcing NVIDIA DGX Cloud through partnerships with Microsoft Azure, Google GCP, and Oracle OCI\u00a0to bring NVIDIA DGX AI supercomputers to every company, instantly, from a browser.\u00a0DGX Cloud is optimized to run NVIDIA AI Enterprise, the world\u00a0's leading acceleration library suite\u00a0for end-to-end development and deployment of AI.\u00a0DGX Cloud offers customers the best of NVIDIA AI and the best of the world\u00a0's leading cloud service providers.\u00a0This partnership brings NVIDIA\u00a0's ecosystem to the CSPs,\u00a0while amplifying NVIDIA\u00a0's scale and reach.\u00a0This win-win partnership gives customers racing to engage Generative AI instant access to NVIDIA in global-scale clouds.\u00a0We\u00a0're excited by the speed, scale, and reach of this cloud extension of our business model.\u00a0Oracle Cloud Infrastructure, OCI,\u00a0will be the first NVIDIA DGX Cloud.\u00a0OCI has excellent performance. They have a two-tier\u00a0computing fabric and management network.\u00a0NVIDIA\u00a0's CX-7, with the industry\u00a0's best RDMA,\u00a0is the computing fabric.\u00a0And BlueField-3 will be the infrastructure processor\u00a0for the management network.\u00a0The combination is a state-of-the-art DGX AI supercomputer that can be offered as a multi-tenant cloud service.\u00a0We have 50 early access enterprise customers, spanning consumer internet and software, healthcare\u00a0media and entertainment, and financial services.\u00a0ChatGPT, Stable Diffusion, DALL-E, and Midjourney have awakened the world to Generative AI.\u00a0These applications\u00a0'\u00a0ease-of-use and impressive capabilities attracted over a hundred million users in just a few months\u00a0- ChatGPT is the fastest-growing application in history.\u00a0No training is necessary. Just ask these models to do something.\u00a0The prompts can be precise or ambiguous. If not clear,\u00a0through conversation, ChatGPT learns your intentions.\u00a0The generated text is beyond impressive.\u00a0ChatGPT can compose memos and poems, paraphrase a research paper, solve math problems,\u00a0highlight key points of a contract,\u00a0and even code software programs.\u00a0ChatGPT is a computer that not only\u00a0runs software but writes software.\u00a0Many breakthroughs led to Generative AI.\u00a0Transformers learn context and meaning from the relationships and dependencies of data, in parallel and at large scale.\u00a0This led to large language models that learn from so much data\u00a0they can perform downstream tasks without explicit training.\u00a0And diffusion models, inspired by physics, learn without\u00a0supervision to generate images.\u00a0In just over a decade, we went from trying to recognize\u00a0cats to generating realistic images of a cat\u00a0in a space suit\u00a0walking on the moon.\u00a0Generative AI is a new kind of computer \u2014 one that we program in human language.\u00a0This ability has profound implications. Everyone can direct\u00a0a computer to solve problems.\u00a0This was a domain only for computer programmers.\u00a0Now everyone is a programmer.\u00a0Generative AI is a new computing platform like PC,\u00a0internet, mobile, and cloud.\u00a0And like in previous computing eras,\u00a0first-movers are creating new applications\u00a0and founding new companies to capitalize on\u00a0Generative AI\u00a0's ability to automate and co-create.\u00a0Debuild lets users design and deploy web applications\u00a0just by explaining what they want.\u00a0Grammarly is a writing assistant that considers context.\u00a0Tabnine helps developers write code.\u00a0Omnekey generates customized ads and copy.\u00a0Kore.ai is a virtual customer service agent.\u00a0Jasper generates marketing material. Jasper has\u00a0written nearly 5 billion words,\u00a0reducing time to generate the first draft by 80%.\u00a0Insilico uses AI to accelerate drug design.\u00a0Absci is using AI to predict therapeutic antibodies.\u00a0Generative AI will reinvent nearly every industry.\u00a0Many companies can use one of the excellent\u00a0Generative AI APIs coming to market.\u00a0Some companies need to build custom models, with their proprietary data, that are experts in their domain.\u00a0They need to set up usage guardrails\u00a0and refine their models to align\u00a0with their company\u00a0's safety, privacy, and security requirements.\u00a0The industry needs a foundry, a TSMC,\u00a0for custom large language models.\u00a0Today, we announce the NVIDIA AI Foundations\u00a0a cloud service for customers needing to build, refine, and operate\u00a0custom LLMlarge language models and Generative AI\u00a0trained with their proprietary data\u00a0and for their domain-specific tasks.\u00a0NVIDIA AI Foundations comprises Language,\u00a0Visual, and Biology model-making services.\u00a0NVIDIA Nemo is for building custom language text-to-text\u00a0generative models.\u00a0Customers can bring their model or start with the Nemo pre-trained language models, ranging from GPT-8, GPT-43\u00a0and GPT-530 billion parameters.\u00a0Throughout the entire process, NVIDIA AI experts will work with you, from creating your proprietary model to operations.\u00a0Let\u00a0's take a look.\u00a0Generative models, like NVIDIA\u00a0's 43B foundational model, learn by training on billions of sentences\u00a0and trillions of words.\u00a0As the model converges, it begins to understand the relationships between words and their underlying concepts\u00a0captured in the weights in the embedding space of the model.\u00a0Transformer models use a technique called self attention: a mechanism designed to learn dependencies and relationships\u00a0within a sequence of words.\u00a0The result is a model that provides the foundation for a ChatGPT-like experience.\u00a0These generative models require expansive amounts of data\u00a0deep AI expertise for data processing and distributed training\u00a0\u00a0and large scale compute to train, deploy\u00a0and maintain at the pace of innovation.\u00a0Enterprises can fast-track their generative AI adoption\u00a0with NVIDIA NeMo service running on NVIDIA DGX Cloud.\u00a0The quickest path is starting with one of NVIDIA\u00a0's state-of-the-art\u00a0pre-trained foundation models.\u00a0With the NeMo service, organizations can easily customize a model\u00a0with p-tuning to teach it specialized skills\u00a0like summarizing financial documents\u00a0creating brand-specific content\u00a0and composing emails with personalized writing styles.\u00a0Connecting the model to a proprietary knowledge base\u00a0ensures that responses are accurate, current\u00a0and cited for their business.\u00a0Next, they can provide guardrails by adding logic\u00a0and monitoring inputs, outputs, toxicity, and bias thresholds\u00a0so it operates within a specified domain\u00a0and prevents undesired responses.\u00a0After putting the model to work, it can continuously improve\u00a0with reinforcement learning based on user interactions.\u00a0And NeMo\u00a0's playground is available for rapid prototyping\u00a0before moving to the cloud API\u00a0for larger-scale evaluation and application integration.\u00a0Sign up for the NVIDIA NeMo service today\u00a0to codify your enterprise\u00a0's knowledge into a personalized\u00a0AI model that you control.\u00a0Picasso is a visual language model-making service for customers who want to build custom models\u00a0trained with licensed or proprietary content.\u00a0Let\u00a0's take a look.\u00a0Generative AI is transforming how visual content is created.\u00a0But to realize its full potential, enterprises need massiveamounts of copyright-cleared data, AI experts, and an AI supercomputer.\u00a0NVIDIA Picasso is a cloud service for building and deploying\u00a0generative AI-powered image, video, and 3D applications.\u00a0With it, enterprises, ISVs, and service providers\u00a0can deploy their own models.\u00a0We're working with premier partners to bring\u00a0generative AI capabilities to every industry\u00a0Organizations can also start with NVIDIA Edify models\u00a0\u00a0and train them on their data to create a product or service.\u00a0These models generate images, videos, and 3D assets.\u00a0To access generative AI models\u00a0applications send an API call with text prompts\u00a0and metadata to Picasso.\u00a0Picasso uses the appropriate model running on NVIDIA DGX Cloud\u00a0to send back the generated asset to the application.\u00a0This can be a photorealistic image, a high-resolution video,\u00a0or a detailed 3D geometry.\u00a0Generated assets can be imported into editing tools or into NVIDIA Omniverse to build photorealistic virtual worlds,\u00a0metaverse applications, and digital twin simulations.\u00a0With NVIDIA Picasso services running on NVIDIA DGX Cloud\u00a0you can streamline training, optimization, and inference\u00a0needed to build custom generative AI applications.\u00a0See how NVIDIA Picasso can bring transformative generative AI capabilities into your applications.\u00a0We are delighted that Getty Images will use the Picasso service to build Edify-image and Edify-video generative models\u00a0trained on their rich library of responsibly licensed\u00a0professional images and video assets.\u00a0Enterprises will be able to create custom images\u00a0and video with simple text or image prompts.\u00a0Shutterstock is developing an Edify-3D generative model\u00a0trained on their professional image, 3D, and video assets library.\u00a0Shutterstock will help simplify the creation of 3D assets for creative production, digital twins and virtual collaboration,\u00a0making these workflows faster\u00a0and easier for enterprises to implement.\u00a0And I\u00a0'm thrilled to announce a significant expansion\u00a0of our long-time partnership with Adobe\u00a0to build a set of next-generation AI capabilities\u00a0for the future of creativity\u00a0integrating generative AI into the everyday workflows\u00a0of marketers and creative professionals.\u00a0The new Generative AI models will be optimized\u00a0for image creation, video, 3D, and animation.\u00a0To protect artists\u00a0'\u00a0rights, Adobe is developing with a focus on commercial viability and proper content attribution\u00a0powered by Adobe\u00a0's Content Authenticity Initiative.\u00a0Our third language domain is biology.\u00a0Drug discovery is a nearly $2T industry\u00a0with $250B dedicated to R&D.\u00a0NVIDIA\u00a0's Clara is a healthcare application framework for imaging\u00a0instruments, genomics, and drug discovery.\u00a0The industry is now jumping onto generative AI\u00a0to discover disease targets\u00a0design novel molecules or protein-based drugs, and predict the behavior of the medicines in the body.\u00a0Insilico Medicine, Exscientia, Absci, and Evozyme, are among hundreds of new AI drug discovery start-ups.\u00a0Several have discovered novel targets or drug candidates and have started human clinical trials.\u00a0BioNeMo helps researchers create\u00a0fine-tune, and serve custom models with their proprietary data.\u00a0Let\u00a0's take a look.\u00a0There are 3 key stages to drug discovery\u00a0discovering the biology that causes disease\u00a0designing new molecules - whether those are small-molecules, proteins or antibodies\u00a0and finally screening how those molecules interact with each other.\u00a0Today, Generative AI is transforming every step of the drug discovery process.\u00a0NVIDIA BioNeMo Service provides state-of-the-art\u00a0generative AI models for drug discovery.\u00a0It\u00a0's available as a cloud service, providing instant and easy access to accelerated drug discovery workflows.\u00a0BioNeMo includes models like AlphaFold, ESMFold and OpenFold\u00a0for 3D protein structure prediction.\u00a0ProtGPT for protein generation,\u00a0ESM1 and ESM2 for protein property prediction\u00a0MegaMolBART and MoFlow and for molecule generation\u00a0and DiffDock for molecular docking.\u00a0Drug discovery teams can use the\u00a0models through BioNeMo\u00a0's web interface\u00a0or cloud APIs.\u00a0Here is an example of using NVIDIA BioNeMo\u00a0for drug discovery virtual screening.\u00a0Generative models can now read a proteins amino acid sequence\u00a0and in seconds, accurately predict the structure of a target protein.\u00a0They can also generate molecules with desirable ADME properties that optimize how a drug behaves in the body.\u00a0Generative models can even predict the 3D interactions\u00a0of a protein and molecule\u00a0accelerating the discovery of optimal drug candidates.\u00a0With NVIDIA DGX Cloud BioNeMo also provides on-demand super computing infrastructure to further optimize and train models,\u00a0saving teams valuable time and money so they can focus on discovering life saving medicines.\u00a0The new AI drug discovery pipelines are here.\u00a0Sign up for access for NVIDIA BioNeMo Service.\u00a0We will continue to work with the industry\u00a0to include models into BioNemo\u00a0that encompass the end-to-end workflow of\u00a0drug discovery and virtual screening.\u00a0Amgen, AstraZeneca, Insilico Medicine, Evozyne, Innophore, and Alchemab Therapeutics are early access users of BioNeMo.\u00a0NVIDIA AI Foundations, a cloud service, a foundry, for building custom language models and Generative AI.\u00a0Since AlexNet a decade ago, deep learning has opened giant new markets \u2014 automated driving, robotics, smart speakers,\u00a0and reinvented how we shop, consume news, and enjoy music.\u00a0That\u00a0's just the tip of the iceberg.\u00a0AI is at an inflection point as Generative AI has started a new wave of opportunities, driving a step-function increase\u00a0in inference workloads.\u00a0AI can now generate diverse data, spanning voice, text, images, video, and 3D graphics to proteins and chemicals.\u00a0Designing a cloud data center to\u00a0process Generative AI is a great challenge.\u00a0On the one hand, a single type of accelerator is ideal,\u00a0because it allows the datacenter to be elastic\u00a0and handle the unpredictable peaks and valleys of traffic.\u00a0On the other hand, no one accelerator can optimally process the diversity of algorithms, models, data types, and sizes.\u00a0NVIDIA's One Architecture platform\u00a0offers both acceleration and elasticity.\u00a0Today, we are announcing our new inference platform - four configurations - one architecture - one software stack.\u00a0Each configuration is optimized for a class of workloads.\u00a0For AI video workloads, we have L4 optimized for video decoding and transcoding, video content moderation,\u00a0and video call features like background replacement,\u00a0relighting, making eye contact,\u00a0transcription, and real-time language translation.\u00a0Most cloud videos today are processed on CPUs.\u00a0One 8-GPU L4 server will replace over a hundred dual-socket CPU servers for processing AI Video.\u00a0Snap is a leading user of NVIDIA AI for computer vision\u00a0and recommender systems.\u00a0Snap will use L4 for AV1 video processing,\u00a0generative AI, and augmented reality.\u00a0Snapchat users upload hundreds of millions of videos every day.\u00a0Google announced today NVIDIA L4 on GCP.\u00a0NVIDIA and Google Cloud are working\u00a0to deploy major workloads on L4.\u00a0Let me highlight five.\u00a0First, we\u00a0're accelerating inference for generative AI models for cloud services like Wombo and Descript.\u00a0Second, we\u00a0're integrating Triton Inference Server with Google Kubernetes Engine and VertexAI.\u00a0Third, we\u00a0're accelerating Google Dataproc\u00a0with NVIDIA Spark-RAPIDS.\u00a0Fourth, we\u00a0're accelerating AlphaFold,\u00a0and UL2 and T5 large language models.\u00a0And fifth, we are accelerating Google Cloud\u00a0's Immersive Stream that renders 3D and AR experiences.\u00a0With this collaboration, Google GCP is a premiere NVIDIA AI cloud.\u00a0We look forward to telling you even more\u00a0about our collaboration very soon.\u00a0For Omniverse, graphics rendering and generative AI like text-to-image and text-to-video, we are announcing L40.\u00a0L40 is up to 10 times the performance of NVIDIA\u00a0's T4,\u00a0the most popular cloud inference GPU.\u00a0Runway is a pioneer in Generative AI.\u00a0Their research team was a key creator of Stable Diffusion\u00a0and its predecessor, Latent Diffusion.\u00a0Runway is inventing generative AI models\u00a0for creating and editing content.\u00a0With over 30 AI Magic Tools, their service is revolutionizing the creative process, all from the cloud.\u00a0Let's take a look.\u00a0Runway is making amazing AI-powered video editing and image creation tools accessible to everyone.\u00a0Powered by the latest generation of NVIDIA GPUs running locally or in the cloud, Runway makes it possible\u00a0\u00a0to remove an object from a video with just a few brush strokes.\u00a0Or apply different styles to video using just an input image.\u00a0Or change the background or the foreground of a video.\u00a0What used to take hours using conventional tools can now be completed with professional broadcast quality results\u00a0in just a few minutes.\u00a0Runway does this by utilizing CV-CUDA, an open-source project that enables developers to build highly efficient\u00a0GPU-accelerated pre- and post-processing pipelines for computer vision workloads and scale them into the cloud.\u00a0With NVIDIA technology, Runway is able to make impossible things to give the best experience to content creators.\u00a0What previously limited pros can now be done by you.\u00a0In fact, Runway is used in Oscar-nominated Hollywood films and we are placing this technology\u00a0in the hands of the world's creators.\u00a0Large language models like ChatGPT\u00a0are a significant new inference workload.\u00a0GPT models are memory and computationally intensive.\u00a0Furthermore, inference is a high-volume, scale-out workload and requires standard commodity servers.\u00a0For large language model inference, like ChatGPT, we are announcing a new Hopper GPU \u2014 the PCIE H100\u00a0with dual-GPU NVLINK.\u00a0The new H100 has 94GB of HBM3 memory.\u00a0H100 can process the 175-billion-parameter GPT-3\u00a0and supporting commodity PCIE servers make it easy to scale out.\u00a0The only GPU in the cloud today that can practically process ChatGPT is HGX A100.\u00a0Compared to HGX A100 for GPT-3 processing, a standard server with four pairs of H100 with dual-GPU NVLINK\u00a0\u00a0is up to 10X faster.\u00a0H100 can reduce large language model processing costs\u00a0by an order of magnitude.\u00a0Grace Hopper is our new superchip that connects Grace CPU and Hopper GPU over a high-speed 900 GB\/sec\u00a0coherent chip-to-chip interface.\u00a0Grace Hopper is ideal for processing giant data sets like AI databases for recommender systems\u00a0and large language models.\u00a0Today, CPUs, with large memory, store and query giant embedding tables then transfer results to GPUs for inference.\u00a0With Grace-Hopper, Grace queries the embedding tables and transfers the results directly to Hopper\u00a0across the high-speed interface \u2013 7 times faster than PCIE.\u00a0Customers want to build AI databases\u00a0several orders of magnitude larger.\u00a0Grace-Hopper is the ideal engine.\u00a0This is NVIDIA's inference platform \u2013\u00a0one architecture for diverse AI workloads,\u00a0\u00a0and maximum datacenter acceleration and elasticity.\u00a0The world\u00a0's largest industries make physical things,\u00a0but they want to build them digitally.\u00a0Omniverse is a platform for industrial digitalization\u00a0that bridges digital and physical.\u00a0It lets industries design, build, operate, and optimize physical products and factories digitally,\u00a0before making a physical replica.\u00a0Digitalization boosts efficiency and speed and saves money.\u00a0One use of Omniverse is the virtual bring-up of a factory, where all of its machinery is integrated digitally\u00a0before the real factory is built.\u00a0This reduces last-minute surprises, change orders,\u00a0and plant opening delays.\u00a0Virtual factory integration can save billions for the world\u00a0's factories.\u00a0The semiconductor industry is investing half a trillion dollars\u00a0to build a record 84 new fabs.\u00a0By 2030, auto manufacturers will build 300 factories\u00a0to make 200 million electric vehicles.\u00a0And battery makers are building 100 more mega factories.\u00a0Digitalization is also transforming logistics, moving goods through billions of square feet of warehouses worldwide.\u00a0Let\u00a0's look at how Amazon uses Omniverse to automate, optimize, and plan its autonomous warehouses.\u00a0Amazon Robotics has manufactured and deployed the largest fleet of mobile industrial robots in the world.\u00a0The newest member of this robotic fleet is Proteus, Amazon's first fully autonomous warehouse robot.\u00a0Proteus is built to move through our facilities using advanced safety, perception, and navigation technology.\u00a0Let's see how NVIDIA Isaac Sim, built on Omniverse is creating physically accurate, photoreal simulations\u00a0to help accelerate Proteus deployments.\u00a0Proteus features multiple sensors that include cameras,\u00a0lidars, and ultrasonic sensors\u00a0to power it\u00a0's autonomy software systems.\u00a0The Proteus team needed to improve the performance of a neural network that read fiducial markers and helped the robot\u00a0determine its location on the map.\u00a0It takes lots of data\u2014and the right kind\u2014to train the ML models that are driven by the robot sensor input.\u00a0With Omniverse Replicator in Isaac Sim, Amazon Robotics was able to generate large photoreal synthetic datasets that improved\u00a0the marker detection success rate from 88.6% to 98%.\u00a0The use of the synthetic data generated by Omniverse Replicator also sped up development times, from months to days,\u00a0as we were able to iteratively test and train our models\u00a0much faster than when only using real data.\u00a0To enable new autonomous capabilities for the expanding fleet of Proteus robots, Amazon Robotics is working towards\u00a0closing the gap from simulation to reality, building large scale multi-sensor, multi-robot simulations.\u00a0With Omniverse, Amazon Robotics will optimize operations with full fidelity warehouse digital twins.\u00a0Whether we're generating synthetic data or developing new levels of autonomy, Isaac Sim on Omniverse\u00a0\u00a0helps the Amazon Robotics team save time and money as we deploy Proteus across our facilities.\u00a0Omniverse has unique technologies for digitalization.\u00a0And Omniverse is the premier development platform for USD, which serves as a common language that lets teams collaborate\u00a0to create virtual worlds and digital twins.\u00a0Omniverse is physically based, mirroring the laws of physics.\u00a0It can connect to robotic systems and operate with hardware-in-the-loop.\u00a0It features Generative AI to accelerate the creation of virtual worlds.\u00a0And Omniverse can manage data sets of enormous scale.\u00a0We've made significant updates to Omniverse in every area.\u00a0Let\u00a0's take a look.\u00a0Nearly 300,000 creators and designers\u00a0have downloaded Omniverse.\u00a0Omniverse is not a tool, but a USD network and shared database,\u00a0a fabric connecting to design tools used across industries.\u00a0It connects, composes, and simulates the assets\u00a0created by industry-leading tools.\u00a0We are delighted to see the growth of Omniverse connections.\u00a0Each connection links the ecosystem of one platform to the ecosystems of all the others.\u00a0Omniverse\u00a0's network of networks is growing exponentially.\u00a0Bentley Systems LumenRT is now connected.\u00a0So are Siemens Teamcenter, NX, and Process Simulate, Rockwell Automation Emulate 3D, Cesium, Unity, and many more.\u00a0Let\u00a0's look at the digitalization of the $3T auto industry\u00a0and see how car companies are\u00a0evaluating Omniverse in their workflows.\u00a0Volvo Cars and GM use Omniverse USD Composer\u00a0to connect and unify their asset pipelines.\u00a0GM connects designers, sculptors, and artists using Alias, Siemens NX, Unreal, Maya, 3ds Max,\u00a0and virtually assembles the components\u00a0into a digital twin of the car.\u00a0\u00a0In engineering and simulation, they visualize the power flow aerodynamics in Omniverse.\u00a0For next-generation Mercedes-Benz and Jaguar Land Rover vehicles, engineers use Drive Sim in Omniverse to generate\u00a0synthetic data to train AI models, validate the active-safety system against a virtual NCAP driving test,\u00a0and simulate real driving scenarios.\u00a0Omniverse\u00a0's generative AI reconstructs\u00a0previously driven routes into 3D\u00a0\u00a0so past experiences can be reenacted or modified.\u00a0Working with Idealworks, BMW uses Isaac Sim\u00a0in Omniverse to generate synthetic data\u00a0and scenarios to train factory robots.\u00a0Lotus is using Omniverse to virtually assemble welding stations.\u00a0Toyota is using Omniverse to build digital twins of their plants.\u00a0Mercedes-Benz uses Omniverse to build, optimize, and plan assembly lines for new models.\u00a0Rimac and Lucid Motors use Omniverse to build digital stores from actual design data that faithfully represent their cars.\u00a0BMW is using Omniverse to plan operations across nearly three dozen factories worldwide.\u00a0And they are building a new EV factory, completely in Omniverse, two years before the physical plant opens.\u00a0Let's visit.\u00a0The world\u00a0's industries are accelerating digitalization with over $3.4 trillion being invested in the next three years.\u00a0We at BMW strive to be leading edge in automotive digitalization.\u00a0With NVIDIA Omniverse and AI we set up new factories faster and produce more efficiently than ever.\u00a0This results in significant savings for us.\u00a0It all starts with planning \u2013 a complex process\u00a0in which we need to connect many tools,\u00a0datasets and specialists around the world.\u00a0Traditionally, we are limited, since data is managed separately in a variety of systems and tools.\u00a0Today, we\u00a0've changed all that.\u00a0We are developing custom Omniverse applications to connect our existing tools, know-how and teams\u00a0all in a unified view.\u00a0Omniverse is cloud-native and cloud-agnostic enabling teams to collaborate across our virtual factories from everywhere.\u00a0I\u00a0'm about to join a virtual planning session for Debrecen in Hungary \u2013 our new EV factory \u2013 opening in 2025.\u00a0Let\u02bcs jump in.\u00a0Planner 1: Ah, Milan is joining.\u00a0Milan: Hello, everyone!\u00a0Planner 1:Hi Milan \u2013 great to see you, we\u00a0're in the middle of an optimization loop for our body shop.\u00a0Would you like to see?\u00a0Milan: Thanks \u2013 I\u00a0'm highly interested. And I\u00a0'd like to invite a friend.\u00a0Planner 1: Sure.\u00a0Jensen: Hey Milan! Good to see you.\u00a0Milan: Jensen, welcome to our virtual planning session.\u00a0Jensen: Its great to be here. What are we looking at?\u00a0Milan: This is our global planning team who are working on a robot cell in Debrecen\u00a0's digital twin.\u00a0Matthias, tell us what\u00a0's happening \u2026\u00a0Matthias: So, we just learned the\u00a0production concept requires some changes.\u00a0We\u00a0're now reconfiguring the layout to add a new robot into the cell.\u00a0Planner 2: Ok, but if we add a new robot, on the logistics side, we\u00a0'll need to move our storage container.\u00a0Planner 3: Alright, let's get this new robot in.\u00a0Matthias: That\u00a0's perfect. But let\u00a0's double-check -\u00a0can we run the cell?\u00a0Excellent.\u00a0Jensen: Milan, this is just incredible!\u00a0Virtual factory integration is essential for every industry.\u00a0I\u00a0'm so proud to see what our teams did together. Congratulations!\u00a0Milan: We are working globally to optimize locally.\u00a0After planning, operations is king, and we\u00a0've already started!\u00a0To celebrate the launch of our virtual plant, I\u00a0'd like to invite you to open the first digital factory with me.\u00a0Jensen: I\u00a0'd be honored. Let\u00a0's do it!\u00a0Car companies employ nearly 14 million people.\u00a0Digitalization will enhance the industry's\u00a0efficiency, productivity, and speed.\u00a0Omniverse is the digital-to-physical operating system\u00a0to realize industrial digitalization.\u00a0Today we are announcing three\u00a0systems designed to run Omniverse.\u00a0First, we\u00a0're launching a new generation of workstations powered by NVIDIA Ada RTX GPUs and Intel's newest CPUs.\u00a0The new workstations are ideal for doing ray tracing, physics simulation, neural graphics, and generative AI.\u00a0They will be available from Boxx, Dell, HP,\u00a0and Lenovo starting in March.\u00a0Second, new NVIDIA OVX servers optimized for Omniverse.\u00a0OVX consists of L40 Ada RTX server\u00a0GPUs and our new BlueField-3.\u00a0OVX servers will be available from Dell, HPE, Quanta, Gigabyte, Lenovo, and Supermicro.\u00a0Each layer of the Omniverse stack, including the chips, systems, networking, and software are new inventions.\u00a0Building and operating the Omniverse computer\u00a0requires a sophisticated IT team.\u00a0We\u00a0're going to make Omniverse\u00a0fast and easy to scale and engage.\u00a0Let\u00a0's take a look.\u00a0The world\u00a0's largest industries are racing\u00a0to digitalize their physical processes.\u00a0Today, that\u00a0's a complex undertaking.\u00a0NVIDIA Omniverse Cloud is a platform-as-a-service that provides instant, secure access to managed Omniverse Cloud APIs,\u00a0workflows, and customizable applications running on NVIDIA OVX.\u00a0Enterprise teams access the suite of managed services through the web browser Omniverse Launcher\u00a0or via a custom-built integration.\u00a0Once in Omniverse Cloud, enterprise teams can instantly access, extend, and publish foundation applications\u00a0\u00a0and workflows - to assemble and compose virtual worlds -\u00a0generate data to train perception AIs -\u00a0test and validate autonomous vehicles -\u00a0or simulate autonomous robots\u2026\u00a0\u2026accessing and publishing shared data to Omniverse Nucleus.\u00a0Designers and engineers working in their favorite\u00a03rd party design tools on RTX workstations,\u00a0publish edits to Nucleus in parallel.\u00a0Then when ready to iterate or view\u00a0their integrated model in Omniverse,\u00a0can simply open a web browser and log in.\u00a0As projects and teams scale, Omniverse Cloud helps optimize cost\u00a0by provisioning compute resources and licenses as needed.\u00a0And new services and upgrades are automatically\u00a0provided with real time updates.\u00a0With Omniverse Cloud, enterprises can fast-track unified digitalization and collaboration\u00a0across major industrial workflows, increasing efficiency,\u00a0reducing costs and waste,\u00a0\u00a0and accelerating the path to innovation.\u00a0See you in Omniverse!\u00a0Today, we announce the NVIDIA Omniverse Cloud,\u00a0a fully managed cloud service.\u00a0We\u00a0're partnering with Microsoft to bring Omniverse Cloud\u00a0to the world\u00a0's industries.\u00a0We will host it in Azure, benefiting from Microsoft\u00a0's rich storage, security, applications, and services portfolio.\u00a0We are connecting Omniverse Cloud to Microsoft 365 productivity suite, including Teams, OneDrive, SharePoint,\u00a0\u00a0and the Azure IoT Digital Twins services.\u00a0Microsoft and NVIDIA are bringing Omniverse to hundreds of millions of Microsoft 365 and Azure users.\u00a0Accelerated computing and AI have arrived.\u00a0Developers use NVIDIA to speed-up and scale-up to solve problems previously impossible.\u00a0A daunting challenge is Net Zero. Every company must accelerate every workload to reclaim power.\u00a0Accelerated computing is a full-stack,\u00a0datacenter-scale computing challenge.\u00a0Grace, Grace-Hopper, and BlueField-3 are new chips for super energy-efficient accelerated data centers.\u00a0Acceleration libraries solve new challenges and open new markets.\u00a0We updated 100 acceleration libraries, including cuQuantum for quantum computing, cuOpt for combinatorial optimization,\u00a0and cuLitho for computational lithography.\u00a0We are thrilled to partner with TSMC, ASML,\u00a0and Synopsys to go to 2nm and beyond.\u00a0NVIDIA DGX AI Supercomputer is the engine behind the generative large language model breakthrough.\u00a0The DGX H100 AI Supercomputer\u00a0is in production and available soon\u00a0from an expanding network of OEM and cloud partners worldwide.\u00a0The DGX supercomputer is going beyond\u00a0research and becoming a modern AI factory.\u00a0Every company will manufacture intelligence.\u00a0We are extending our business model with NVIDIA DGX Cloud by partnering with Microsoft Azure, Google GCP, and Oracle OCI\u00a0to instantly bring NVIDIA AI to every company, from a browser.\u00a0DGX Cloud offers customers the best of NVIDIA\u00a0and the best of the world\u00a0's leading CSPs.\u00a0We are at the iPhone moment for AI.\u00a0Generative AI inference workloads have gone into overdrive.\u00a0We launched our new inference platform -\u00a0four configurations - one architecture.\u00a0L4 for AI video.\u00a0L40 for Omniverse and graphics rendering.\u00a0H100 PCIE for scaling out large language model inference.\u00a0Grace-Hopper for recommender systems and vector databases.\u00a0NVIDIA\u00a0's inference platform enables maximum\u00a0data center acceleration and elasticity.\u00a0NVIDIA and Google Cloud are working together to deploy a broad range of inference workloads.\u00a0With this collaboration, Google GCP is a premiere NVIDIA AI cloud.\u00a0NVIDIA AI Foundations is a cloud service, a foundry, for building custom language models and Generative AI.\u00a0NVIDIA AI Foundations comprises language,\u00a0visual, and biology model-making services.\u00a0Getty Images and Shutterstock are\u00a0building custom visual language models.\u00a0And we're partnering with Adobe to build a set of next-generation AI capabilities for the future of creativity.\u00a0Omniverse is the digital-to-physical operating system\u00a0to realize industrial digitalization.\u00a0Omniverse can unify the end-to-end workflow and digitalize the $3T, 14 million-employee automotive industry.\u00a0Omniverse is leaping to the cloud.\u00a0Hosted in Azure, we partner with Microsoft to bring Omniverse Cloud to the world\u00a0's industries.\u00a0I thank our systems, cloud, and software partners,\u00a0researchers, scientists,\u00a0and especially our amazing employees\u00a0for building the NVIDIA accelerated computing ecosystem.\u00a0Together, we are helping the world do the impossible.\u00a0Have a great GTC!\u00a0\u00a0<\/p>\n","protected":false},"excerpt":{"rendered":"
GTC 2023 Presentation<\/p>\n","protected":false},"author":1,"featured_media":260,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"_import_markdown_pro_load_document_selector":0,"_import_markdown_pro_submit_text_textarea":"","emotion":"","emotion_color":"","title_style":"","license":"","footnotes":""},"categories":[24],"tags":[27,25,26],"class_list":["post-259","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-24","tag-ai","tag-gtc","tag-nvidia"],"_links":{"self":[{"href":"https:\/\/reverieland.cn\/index.php\/wp-json\/wp\/v2\/posts\/259","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/reverieland.cn\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/reverieland.cn\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/reverieland.cn\/index.php\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/reverieland.cn\/index.php\/wp-json\/wp\/v2\/comments?post=259"}],"version-history":[{"count":1,"href":"https:\/\/reverieland.cn\/index.php\/wp-json\/wp\/v2\/posts\/259\/revisions"}],"predecessor-version":[{"id":261,"href":"https:\/\/reverieland.cn\/index.php\/wp-json\/wp\/v2\/posts\/259\/revisions\/261"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/reverieland.cn\/index.php\/wp-json\/wp\/v2\/media\/260"}],"wp:attachment":[{"href":"https:\/\/reverieland.cn\/index.php\/wp-json\/wp\/v2\/media?parent=259"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/reverieland.cn\/index.php\/wp-json\/wp\/v2\/categories?post=259"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/reverieland.cn\/index.php\/wp-json\/wp\/v2\/tags?post=259"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}