NVIDIA founder and CEO Jensen Huang kicked off the COMPUTEX conference in Taipei today with the first live keynote since the pandemic, as companies embrace the historic wave of generative AI that will transform industries from advertising to manufacturing to telecommunications. Announced a platform that can be used to ride.
“I’m back,” Huang exclaimed on stage after years of virtual keynotes, partly from his own kitchen. “I haven’t given a public speech in almost four years. Good luck!”
Speaking to a packed audience of approximately 3,500 people for nearly two hours, he described accelerated computing services, software, and systems that enable new business models and make current business models more efficient.
“Accelerated computing and AI represent the reinvention of computing,” Huang said. Fan’s trip to his hometown over the past week has been tracked daily by local media.
To demonstrate its power, he used the giant 8K wall we talked about earlier to display a text prompt that generated the theme song for the keynote. It can be sung like a karaoke song. Mr. Huang briefly led the audience in singing the new national anthem, occasionally teasing the crowd in his native Taiwanese.
“We are now at a tipping point in a new computing era, with high-speed computing and AI being adopted by nearly every computing and cloud company in the world,” he said. He said companies and 15,000 startups use NVIDIA technology, 25 of which use NVIDIA technology. Last year alone he had millions of downloads of his CUDA software.
Top news announcements from keynotes
A new engine for enterprise AI
Launched DGX GH200, a high-capacity AI supercomputer for enterprises that need the ultimate in AI performance. Combine up to 256 NVIDIA GH200 Grace Hopper superchips into a single datacenter-sized GPU using NVIDIA NVLink.
According to Jensen, the GH200 superchip is currently in full production, combining the energy-efficient NVIDIA Grace CPU and the high-performance NVIDIA H100 Tensor Core GPU into one superchip.
The DGX GH200 features exaflops of performance and 144 terabytes of shared memory, nearly 500 times more than a single NVIDIA DGX A100 320GB system. It enables developers to build large-scale language models for generative AI chatbots, complex algorithms for recommender systems, and graph neural networks used for fraud detection and data analysis.
Google Cloud, Meta, and Microsoft are expected to be the first to gain access to the DGX GH200, which can be used as a blueprint for future hyperscale generative AI infrastructure.
“The DGX GH200 AI supercomputer integrates NVIDIA’s cutting-edge accelerated computing and networking technologies to expand the frontiers of AI,” Huang told the audience in Taipei. Many of the audience had lined up outside the hall for hours before the doors opened.
NVIDIA is building its own massive AI supercomputer, NVIDIA Helios, which will go live this year. Using four of his DGX GH200 systems linked with NVIDIA Quantum-2 InfiniBand networking to power data throughput for training large-scale AI models.
The DGX GH200 forms the top of hundreds of systems announced at the event. Together, the two companies bring generative AI and accelerated computing to millions of users.
Looking at the big picture, Huang announced that more than 400 system configurations will hit the market powered by NVIDIA’s latest Hopper, Grace, Ada Lovelace and BlueField architectures. They aim to tackle the most complex challenges in AI, data science and high performance computing.
Acceleration at any size
To meet the needs of data centers of all sizes, Huang introduced NVIDIA MGX, a modular reference architecture for creating accelerated servers. It allows system manufacturers to quickly and cost-effectively build over 100 different server configurations to suit a wide range of AI, HPC and NVIDIA Omniverse applications.
MGX allows manufacturers to build CPUs and accelerated servers using a common architecture and modular components. Supports NVIDIA’s full line of GPUs, CPUs, data processing units (DPUs), and network adapters, as well as x86 and Arm processors in a variety of air- and liquid-cooled chassis.
QCT and Supermicro will bring the MGX design to market for the first time in August. Supermicro’s ARS-221GL-NR system, announced at COMPUTEX, uses the Grace CPU, while QCT’s S74G-2U system, also announced at the event, uses the Grace Hopper.
ASRock Rack, ASUS, GIGABYTE, and Pegatron also plan to use MGX to create their next-generation accelerated computers.
5G/6G Calls Grace Hopper
Separately, Huang said NVIDIA is helping shape the future of 5G and 6G wireless and video communications. The demo showed how the AI running on Grace Hopper transformed his 2D video call today into a more lifelike 3D experience, delivering amazing immersion.
Laying the groundwork for a new kind of service, Huang announced that NVIDIA is working with telecommunications giant SoftBank to build a distributed network of data centers in Japan. Offers 5G services and generative AI applications on a common cloud his platform.
The data center uses NVIDIA GH200 superchips and NVIDIA BlueField-3 DPUs in modular MGX systems and NVIDIA Spectrum Ethernet switches to provide the precision timing required for 5G protocols. The platform reduces costs by increasing spectral efficiency while reducing energy consumption.
The system will help SoftBank explore 5G applications in autonomous driving, AI factories, augmented and virtual reality, computer vision and digital twins. Future applications may also include 3D video conferencing and holographic communications.
cloud network turbocharger
Separately, Huang announced NVIDIA Spectrum-X, a networking platform aimed at improving the performance and efficiency of Ethernet-based AI clouds. Combining Spectrum-4 Ethernet switches with his BlueField-3 DPU and software, he improves AI performance and power efficiency by 1.7x compared to traditional Ethernet fabrics.
NVIDIA Spectrum-X, Spectrum-4 switches, and BlueField-3 DPUs are available now from system manufacturers such as Dell Technologies, Lenovo, and Supermicro.
Bring game characters to life
Generative AI will also affect how people play.
Huang announced NVIDIA Avatar Cloud Engine (ACE) for Gaming, a foundry service that developers can use to build and deploy custom AI models for speech, dialogue, and animation. This gives non-playable characters conversational skills, allowing them to respond to questions with an evolving lifelike personality.
NVIDIA ACE for games includes AI foundation models such as NVIDIA Riva to detect and transcribe player speech. This text prompts NVIDIA NeMo to generate a customized response animated with NVIDIA Omniverse Audio2Face.
Accelerate Gen AI on Windows
Huang explained how NVIDIA and Microsoft are working together to drive innovation for Windows PCs in the age of generative AI.
New and enhanced tools, frameworks and drivers make it easier for PC developers to develop and deploy AI. For example, the Microsoft Olive toolchain for optimizing and deploying GPU-accelerated AI models and new graphics drivers improve DirectML performance on Windows PCs with NVIDIA GPUs.
This collaboration will power and expand the installed base of 100 million PCs powered by RTX GPUs powered by Tensor Cores that boost the performance of over 400 AI-accelerated Windows apps and games.
Digitalization of the world’s largest industry
Generative AI is also creating new opportunities for the $700 billion digital advertising industry.
For example, WPP, the world’s largest marketing services organization, is working with NVIDIA to build the first-of-its-kind AI-enabled content engine on Omniverse Cloud.
In a demo, Huang showed how creative teams can connect 3D design tools like Adobe Substance 3D to build digital twins of client products on NVIDIA Omniverse. You can then rapidly create virtual sets using content from generative AI tools trained on responsibly sourced data and built on NVIDIA Picasso. A WPP client can use a complete scene to generate a multitude of ads, videos and 3D experiences for global markets and users to experience on any of her web devices.
“Today we get ads, but in the future, when users access information, a lot of it will be generated. The computing model has changed,” said Huang.
Factories build the future of AI
With an estimated 10 million factories, the $46 trillion manufacturing sector is a rich area of industrial digitalization.
“The world’s largest industries manufacture physical things. Building digital first would save billions of dollars,” Huang said.
Keynotes will showcase how electronics manufacturers such as Foxconn Industrial Internet, Innodisk, Pegatron, Quanta and Wistron are using NVIDIA technology to build digital workflows to realize their vision of an all-digital smart factory. showed.
They use Omniverse and generative AI APIs to connect design and manufacturing tools so they can build a digital twin of their factory. In addition, he uses NVIDIA Isaac Sim for robot simulation and testing and NVIDIA Metropolis, a vision AI framework for automated optical inspection.
Our newest component, NVIDIA Metropolis for Factories, gives manufacturers a competitive edge by creating custom quality management systems. We help companies develop cutting-edge AI applications.
Accelerating Assembly Lines with AI
For example, Pegatron, which manufactures 300 different products worldwide, including laptops and smartphones, is building a virtual factory with Omniverse, Isaac Sim, and Metropolis. This allows you to try out your processes in a simulated environment, saving you time and money.
Pegatron also used the NVIDIA DeepStream software development kit to develop intelligent video applications, resulting in a 10x increase in throughput.
Foxconn Industrial Internet, the service arm of the world’s largest technology manufacturer, is working with NVIDIA Metropolis partners to automate critical aspects of circuit board quality assurance inspection points.
In a video, Huang showed how Techman Robot, a subsidiary of Quanta, leveraged NVIDIA Isaac Sim to optimize inspection of a manufacturing line for a major Taiwan-based company. This is basically using simulated robots to train robots on how to make better robots.
Additionally, Huang unveiled a new platform to enable next-generation Autonomous Mobile Robot (AMR) fleets. The Isaac AMR helps simulate, deploy, and manage fleets of autonomous mobile robots.
A large partner ecosystem, including ADLINK, Aetina, Deloitte, Quantiphi and Siemens, has helped bring all these manufacturing solutions to market, said Huang.
This is another example of how NVIDIA is helping companies realize the benefits of generative AI through accelerated computing.
“I haven’t seen you in a long time, so I have a lot to say,” said the two-hour lecture, which was met with enthusiastic applause.
For more information, read the full keynote.
