gtc conference The Paper Editor Li Yaorui (02:59)
ai (artificial intelligence) chip leader Nvidia CEO Jen-Hsun Huang continues to increase investment in accelerated computing and generative AI.
On March 18, local time, at NVIDIA's largest GTC conference to date, Huang Renxun delivered a keynote speech "Witnessing the Transformational Moment of AI". In addition to introducing the latest AI chip gb200 and other products, Huang Renxun also talked about some prospects for the future of AI, large models and robots.
"Computing Power Demand"
In his speech, Huang Renxun reiterated the point of view he made in previous financial reports: "Accelerated computing has reached a tipping point - general computing has reached a bottleneck."
Less than a month ago, Nvidia released The financial report for the fourth quarter of fiscal year 2024 on January 28 achieved results that far exceeded expectations. Huang Renxun said in the financial report: "Accelerated computing and generative AI have reached a tipping point, and demand from enterprises, industries and countries is surging globally."
Now, Huang Renxun clearly pointed out that the development of general computing is close to the ceiling, and conducted More details: “We need another way to compute so that we can continue to scale, continue to reduce computing costs, and consume more computing resources while remaining sustainable. Accelerated computing is a significant acceleration of general-purpose computing, This is true in every industry."
Huang Renxun explained to the audience that the emergence of multi-modal AI that can handle data under different models has made AI more adaptable and performant. By increasing parameters, these models can handle increasingly complex analyses, but this also means that the demand for computing power will increase significantly, and the need for advanced computing infrastructure will become increasingly urgent.
Huang Renxun said: "Everyone is very familiar with large-scale language models. After the transformer architecture was invented, we expanded the large model at an incredible speed, basically doubling it every six months... If you expand the model Doubling the size means you have twice the brain capacity and need to fill it with twice the information. Whenever you double the number of parameters, you must also increase the number of training tokens appropriately. The combination of these two numbers will It’s the computing scale you need to support.”
At the beginning of his speech, Huang Renxun showed a picture on which the names of companies exhibiting at this GTC were densely listed. Together, these companies form Nvidia's huge ecosystem and almost support the development of global AI. Subsequently, Huang Renxun also introduced a series of "new acceleration ecosystem" partners, including ansys, synopsis, cadence, etc., and mentioned that TSMC and synopsis will put the breakthrough Nvidia computational lithography into production.
Nvidia partners participating in gtc. Source: NVIDIA live broadcast screenshot
"Robot chatgpt moment"
At this point, Huang Renxun introduced the debut of GPUs based on the latest blackwell architecture, and did not forget to emphasize again and again: "We need larger models and larger GPUs." Huang Renxun also Pointing out that "the future is generative", Blackwell GPU is a processor specially created for the generative AI era.
As for the development of the AI software side, with the launch of the new "NVIDIA Inference Microservice (NIM)", industry APIs, AI algorithm support libraries, cloud architecture support, AI algorithm acceleration, customized models and other demands are all injected into a In small "containers", Nvidia's own software ecosystem will continue to be greatly enhanced.
Next, Huang Renxun looked forward to the future of software programming work: "These incredible collections of software will be optimized and packaged (into nim) and put on the website by us. You can download it, carry it, on any cloud or on your own It can be run in a data center without the need to run a supporting workstation... How will we build software in the future? Most likely, you will no longer need to write it from scratch, or write a lot of things similar to python code, but will build a A team composed of AI. You may use a super AI. If you give it a task, it will convert it into an execution task."
In addition, Huang Renxun also talked about a lot, including his prospects for the development of humanoid robot technology.Huang Renxun believes that given that the end-to-end system designed for robots has been developed for a period of time, the "chatgpt moment" for robots may be close at hand.
Musk touted: Nvidia is the best for ai
This packed speech made Huang Renxun jokingly called "Taylor Swift in the AI world" by netizens, and once again reaffirmed Nvidia's reputation as an "AI arms dealer."
Just on Nvidia’s official website news page for releasing Blackwell GPU, it is filled with praises from AI giants about this latest chip, including Sundar Pichai, CEO of Alphabet and Google, Amazon President and CEO Andy Jassy, Dell Technologies founder and CEO Michael Dell, Google DeepMind co-founder and CEO Demis Hassabis, founder of Meta Mark Zuckerberg, CEO, Microsoft Chairman and CEO Satya Nadella, OpenAI CEO Sam Altman, and Oracle Chairman and CTO Larry Ellison, and even Tesla CEO Elon Musk, who has always been vicious, wrote: "There is no hardware now that is more suitable for AI than NVIDIA."
was opened at Huang Renxun's opening After the speech, Microsoft Chairman and CEO Satya Nadella said on the Comprehensive platforms and tools (from chips to software) to build disruptive AI capabilities." Bindu Reddy, CEO of
ai platform company abacus.ai, wrote on the x platform: "Nvidia is the only widely loved company in the world. The company! After all, we must love this company, without them, this AI revolution would not be possible! Thanks to them for solving the innovation problem at rocket speed."
Jim Cramer, a well-known American economic commentator ) also expressed his optimism about NVIDIA: "What NVIDIA has done will jointly create the next industrial revolution, because NVIDIA represents the speed at which we can obtain smarter and better solutions."
