In his first dwell keynote because the pandemic, NVIDIA founder and CEO Jensen Huang right now kicked off the COMPUTEX convention in Taipei, asserting platforms that firms can use to trip a historic wave of generative AI that’s remodeling industries from promoting to manufacturing to telecom.
“We’re again,” Huang roared as he took the stage after years of digital keynotes, some from his house kitchen. “I haven’t given a public speech in virtually 4 years — want me luck!”
Talking for practically two hours to a packed home of some 3,500, he described accelerated computing providers, software program and techniques which are enabling new enterprise fashions and making present ones extra environment friendly.
“Accelerated computing and AI mark a reinvention of computing,” mentioned Huang, whose travels in his hometown over the previous week have been tracked every day by native media.
In an illustration of its energy, he used the large 8K wall he spoke in entrance of to indicate a textual content immediate producing a theme music for his keynote, singable as any karaoke tune. Huang, who sometimes bantered with the gang in his native Taiwanese, briefly led the viewers in singing the brand new anthem.
“We’re now on the tipping level of a brand new computing period with accelerated computing and AI that’s been embraced by virtually each computing and cloud firm on this planet,” he mentioned, noting 40,000 massive firms and 15,000 startups now use NVIDIA applied sciences with 25 million downloads of CUDA software program final yr alone.
High Information Bulletins From the Keynote
A New Engine for Enterprise AI
For enterprises that want the last word in AI efficiency, he unveiled DGX GH200, a large-memory AI supercomputer. It makes use of NVIDIA NVLink to mix as much as 256 NVIDIA GH200 Grace Hopper Superchips right into a single data-center-sized GPU.
The GH200 Superchip, which Huang mentioned is now in full manufacturing, combines an energy-efficient NVIDIA Grace CPU with a high-performance NVIDIA H100 Tensor Core GPU in a single superchip.
The DGX GH200 packs an exaflop of efficiency and 144 terabytes of shared reminiscence, practically 500x greater than in a single NVIDIA DGX A100 320GB system. That lets builders construct massive language fashions for generative AI chatbots, complicated algorithms for recommender techniques, and graph neural networks used for fraud detection and information analytics.
Google Cloud, Meta and Microsoft are among the many first anticipated to realize entry to the DGX GH200, which can be utilized as a blueprint for future hyperscale generative AI infrastructure.
“DGX GH200 AI supercomputers combine NVIDIA’s most superior accelerated computing and networking applied sciences to increase the frontier of AI,” Huang informed the viewers in Taipei, a lot of whom had lined up outdoors the corridor for hours earlier than the doorways opened.
NVIDIA is constructing its personal large AI supercomputer, NVIDIA Helios, coming on-line this yr. It’ll use 4 DGX GH200 techniques linked with NVIDIA Quantum-2 InfiniBand networking to supercharge information throughput for coaching massive AI fashions.
The DGX GH200 varieties the top of lots of of techniques introduced on the occasion. Collectively, they’re bringing generative AI and accelerated computing to thousands and thousands of customers.
Zooming out to the large image, Huang introduced greater than 400 system configurations are coming to market powered by NVIDIA’s newest Hopper, Grace, Ada Lovelace and BlueField architectures. They goal to deal with probably the most complicated challenges in AI, information science and excessive efficiency computing.
Acceleration in Each Measurement
To suit the wants of information facilities of each dimension, Huang introduced NVIDIA MGX, a modular reference structure for creating accelerated servers. System makers will use it to rapidly and cost-effectively construct greater than 100 totally different server configurations to swimsuit a variety of AI, HPC and NVIDIA Omniverse purposes.
MGX lets producers construct CPU and accelerated servers utilizing a typical structure and modular elements. It helps NVIDIA’s full line of GPUs, CPUs, information processing models (DPUs) and community adapters in addition to x86 and Arm processors throughout a wide range of air- and liquid-cooled chassis.
QCT and Supermicro would be the first to market with MGX designs showing in August. Supermicro’s ARS-221GL-NR system introduced at COMPUTEX will use the Grace CPU, whereas QCT’s S74G-2U system, additionally introduced on the occasion, makes use of Grace Hopper.
ASRock Rack, ASUS, GIGABYTE and Pegatron may also use MGX to create next-generation accelerated computer systems.
5G/6G Requires Grace Hopper
Individually, Huang mentioned NVIDIA helps form future 5G and 6G wi-fi and video communications. A demo confirmed how AI working on Grace Hopper will rework right now’s 2D video calls into extra lifelike 3D experiences, offering a tremendous sense of presence.
Laying the groundwork for brand spanking new sorts of providers, Huang introduced NVIDIA is working with telecom large SoftBank to construct a distributed community of information facilities in Japan. It’ll ship 5G providers and generative AI purposes on a typical cloud platform.
The info facilities will use NVIDIA GH200 Superchips and NVIDIA BlueField-3 DPUs in modular MGX techniques in addition to NVIDIA Spectrum Ethernet switches to ship the extremely exact timing the 5G protocol requires. The platform will scale back value by rising spectral effectivity whereas decreasing power consumption.
The techniques will assist SoftBank discover 5G purposes in autonomous driving, AI factories, augmented and digital actuality, laptop imaginative and prescient and digital twins. Future makes use of might even embody 3D video conferencing and holographic communications.
Turbocharging Cloud Networks
Individually, Huang unveiled NVIDIA Spectrum-X, a networking platform purpose-built to enhance the efficiency and effectivity of Ethernet-based AI clouds. It combines Spectrum-4 Ethernet switches with BlueField-3 DPUs and software program to ship 1.7x beneficial properties in AI efficiency and energy effectivity over conventional Ethernet materials.
NVIDIA Spectrum-X, Spectrum-4 switches and BlueField-3 DPUs can be found now from system makers together with Dell Applied sciences, Lenovo and Supermicro.
Bringing Sport Characters to Life
Generative AI impacts how individuals play, too.
Huang introduced NVIDIA Avatar Cloud Engine (ACE) for Video games, a foundry service builders can use to construct and deploy customized AI fashions for speech, dialog and animation. It’ll give non-playable characters conversational expertise to allow them to reply to questions with lifelike personalities that evolve.
NVIDIA ACE for Video games consists of AI basis fashions corresponding to NVIDIA Riva to detect and transcribe the participant’s speech. The textual content prompts NVIDIA NeMo to generate custom-made responses animated with NVIDIA Omniverse Audio2Face.
Accelerating Gen AI on Home windows
Huang described how NVIDIA and Microsoft are collaborating to drive innovation for Home windows PCs within the generative AI period.
New and enhanced instruments, frameworks and drivers are making it simpler for PC builders to develop and deploy AI. For instance, the Microsoft Olive toolchain for optimizing and deploying GPU-accelerated AI fashions and new graphics drivers will enhance DirectML efficiency on Home windows PCs with NVIDIA GPUs.
The collaboration will improve and prolong an put in base of 100 million PCs sporting RTX GPUs with Tensor Cores that enhance efficiency of greater than 400 AI-accelerated Home windows apps and video games.
Digitizing the World’s Largest Industries
Generative AI can also be spawning new alternatives within the $700 billion digital promoting business.
For instance, WPP, the world’s largest advertising and marketing providers group, is working with NVIDIA to construct a first-of-its sort generative AI-enabled content material engine on Omniverse Cloud.
In a demo, Huang confirmed how artistic groups will join their 3D design instruments corresponding to Adobe Substance 3D, to construct digital twins of consumer merchandise in NVIDIA Omniverse. Then, content material from generative AI instruments educated on responsibly sourced information and constructed with NVIDIA Picasso will allow them to rapidly produce digital units. WPP shoppers can then use the entire scene to generate a number of advertisements, movies and 3D experiences for international markets and customers to expertise on any net gadget.
“At present advertisements are retrieved, however sooner or later if you interact data a lot of it will likely be generated — the computing mannequin has modified,” Huang mentioned.
Factories Forge an AI Future
With an estimated 10 million factories, the $46 trillion manufacturing sector is a wealthy discipline for industrial digitalization.
“The world’s largest industries make bodily issues. Constructing them digitally first can save billions,” mentioned Huang.
The keynote confirmed how electronics makers together with Foxconn Industrial Web, Innodisk, Pegatron, Quanta and Wistron are forging digital workflows with NVIDIA applied sciences to understand the imaginative and prescient of a wholly digital good manufacturing unit.
They’re utilizing Omniverse and generative AI APIs to attach their design and manufacturing instruments to allow them to construct digital twins of factories. As well as, they use NVIDIA Isaac Sim for simulating and testing robots and NVIDIA Metropolis, a imaginative and prescient AI framework, for automated optical inspection.
The newest part, NVIDIA Metropolis for Factories, can create customized quality-control techniques, giving producers a aggressive benefit. It’s serving to firms develop state-of-the-art AI purposes.
AI Speeds Meeting Strains
For instance, Pegatron — which makes 300 merchandise worldwide, together with laptops and smartphones — is creating digital factories with Omniverse, Isaac Sim and Metropolis. That lets it check out processes in a simulated surroundings, saving time and price.
Pegatron additionally used the NVIDIA DeepStream software program improvement package to develop clever video purposes that led to a 10x enchancment in throughput.
Foxconn Industrial Web, a service arm of the world’s largest know-how producer, is working with NVIDIA Metropolis companions to automate vital parts of its circuit-board quality-assurance inspection factors.
In a video, Huang confirmed how Techman Robotic, a subsidiary of Quanta, tapped NVIDIA Isaac Sim to optimize inspection on the Taiwan-based large’s manufacturing traces. It’s primarily utilizing simulated robots to coach robots find out how to make higher robots.
As well as, Huang introduced a brand new platform to allow the following era of autonomous cellular robotic (AMR) fleets. Isaac AMR helps simulate, deploy and handle fleets of autonomous cellular robots.
A big accomplice ecosystem — together with ADLINK, Aetina, Deloitte, Quantiphi and Siemens — helps convey all these manufacturing options to market, Huang mentioned.
It’s yet another instance of how NVIDIA helps firms really feel the advantages of generative AI with accelerated computing.
“It’s been a very long time since I’ve seen you, so I had loads to inform you,” he mentioned after the two-hour discuss to enthusiastic applause.
To be taught extra, try the highlights of NVIDIA’s week at Computex 2023 in a four-minute video right here. And watch the total keynote under
