When large AI models swept the world, Huang Renxun and NVIDIA, who controlled the computing power, unsurprisingly became the biggest winners.
On March 18, the three-day NVIDIA 2024 GTC conference was officially held at the San Jose Convention Center. NVIDIA founder Huang Jenxun wore his iconic leather jacket and delivered a keynote speech “Witness the Transformative Moment of AI.”
GTC (GPU Technology Conference) is a global technology event hosted by NVIDIA, focusing on GPU technology and its application and development in many fields such as high-performance computing, artificial intelligence, deep learning, and financial technology.
The content of the conference is related to the key trends in global AI development and is known as the Spring Festival Gala in the artificial intelligence circle.
In a two-hour speech, founder Huang Jenxun introduced Nvidia’s latest research and development progress around five major sections:
The company’s new industry development
New flagship AI chip Blackwell GPU
New AI microservices NIMs (Nvidia inference micro service)
AI platform NEMO and AI workshop services
Simulation platform Omniverse and robotics platform Isaac Robotics
01、The most powerful AI chip is born
At 13:00 local time, Huang Renxun first reviewed Nvidia’s milestones in technology development over the past 30 years.
Later, Huang Renxun talked about the rapid evolution of AI models that have driven the skyrocketing demand for training computing power and the need for stronger GPUs: “In every industry, accelerated computing has greatly improved compared to general computing.” Then, Huang Renxun dropped the “nuclear bomb” of this speech and announced: “The new flagship AI chip Blackwell GPU is here!”
This chip has 208 billion transistors and its computing power is powerful enough to train a GPT model with 1.8 trillion parameters in 90 days, while consuming only a quarter of the original power. This breakthrough progress not only greatly improves the efficiency of AI model training, but also lays a solid foundation for the widespread application of AI technology in various industries.
Nvidia says Blackwell-based processors like the GB200 offer AI companies a huge performance upgrade, with AI performance of 20 petaflops compared to 4 petaflops for the H100 Operation. The system can deploy a 27 trillion parameter model. Huang Renxun said that there is no doubt that Microsoft Azure, AWS, Google Cloud and other technology giants are the first users of the Blackwell architecture.
02、NIM microservices
In his speech, Huang Renxun also announced NVIDIA’s new progress in the field of AI software-NIM microservices. This new service model will completely change the way software is developed.
In the future, developers may no longer need to write code line by line, but instead mobilize a series of NIMs to complete training and application deployment through communication with AI. NVIDIA’s vision is to become the “fab” for AI software, providing developers around the world with the most powerful and efficient AI production containers.
Currently, NVIDIA’s AI microservices NIM website is online. According to reports, NIM microservices provide the fastest and highest-performance AI production containers for deploying models from NVIDIA, A121, Adept, Cohere, Getty Images and Shutterstock, as well as models from Google, Hugging Face, Meta, Microsoft, Mistral AI and Stability AI’s open model.
Through NIM, enterprises can more efficiently utilize existing GPU resources and reduce the initial training and inference costs of new artificial intelligence models. In addition, NIM software will also support running AI on laptops equipped with GPUs, making AI technology closer to ordinary users and no longer limited to cloud servers.
03、Omniverse和Isaac Robotics
Omniverse is a virtual laboratory where developers can use multiple high-speed computers (GPUs) to simulate and test complex physical systems. The platform’s new capabilities mean we can more realistically simulate real-world physics in a virtual world for designing and testing new products.
Huang Renxun emphasized that Nvidia’s goal is to simulate the entire product in a completely fidelity and completely digital way, thereby expanding the scale of computing. This new simulation tool can not only improve the efficiency of product design, but also test and optimize product performance in a virtual environment.
On the same day, NVIDIA also released a number of updates to Issac. Isaac is an autonomous machine platform for robots launched by NVIDIA in 2018. The platform accelerates the development process of robots through improvements in robot development, simulation and deployment functions. The Issac platform mainly includes hardware, software and virtual world robot simulators.
At the same time, Huang Renxun announced Project GR00T at the press conference, which is a humanoid robot model platform designed to further promote Nvidia’s research and development in robots.
ProjectGR00T is a universal base model for multi-modal humanoid robots that serves as the robot’s brain, enabling them to learn skills to solve a variety of tasks.
GR00T-powered robots will be able to understand natural language and imitate movements by observing human behavior – quickly learning coordination, flexibility and other skills to navigate, adapt to and interact with the real world. In the keynote speech at the GTC conference, NVIDIA founder and CEO Jensen Huang showed how several such robots complete various tasks.
04、Nvidia announced collaborative products
Lenovo: Nvidia and Lenovo announced their cooperation to launch a new hybrid artificial intelligence solution that will provide tailored generative AI. This is a further expansion of the cooperation between the two parties in workstations, gaming PCs and high-performance computers.
Hon Hai Technology Group (Hongbai Technology, a subsidiary of Industrial Fii): demonstrated a new generation of AI servers and liquid refrigerators developed in cooperation with Nvidia at the GTC conference. This shows that Fii’s AI server business in the cloud computing sector is growing.
BYD: Announced that it will use NVIDIA’s next-generation automotive chip Drive Thor, which will help BYD’s technological progress in the field of smart cars.
Xpeng Motors and GAC Aian’s Hyper: Nvidia will expand cooperation with these companies, which may involve the application of Drive Thor technology to improve the performance of autonomous driving and in-car infotainment systems.
Ideal and Geely’s Ji Krypton Auto: have previously announced that they will adopt NVIDIA Drive Thor technology, which will further promote the development of these companies in smart car technology.
Company Nine: Displayed the Isaac Perceptor product jointly developed with NVIDIA, as well as Nova Carter AMR, which is a customizable autonomous driving R&D platform that can be used for the development of autonomous driving cargo robots.
Deloitte: As one of the world’s largest professional services companies, Deloitte is providing new services built on NVIDIA AI and NVIDIA Omniverse to global enterprises, which will help enterprises build modern multi-cloud applications in multiple fields.
Johnson & Johnson Medical Technology: Johnson & Johnson Medical Technology, a world-renowned medical and health company, has cooperated with NVIDIA to jointly develop new artificial intelligence surgical applications, aiming to further improve the accuracy and safety of surgery.
After the conference, as of the close of U.S. stocks on March 19, NVIDIA’s stock price rose by more than 1%, with its latest market value approaching US$2.24 trillion. In the AI era, those with computing power win the world, and there is no doubt that NVIDIA has done it.
The rise of Nvidia is an inevitable trend of the times. The 2024 NVIDIA GTC Conference was silently announced, and a new era has begun.