On June 2nd at 7 PM, Jensen Huang, dubbed the "AI Father" and CEO of NVIDIA, delivered a keynote speech at the National Taiwan University Sports Center as a prelude to the 2024 Taipei International Computer Show (COMPUTEX). During the event, he declared that "a new era of computing is beginning," showcasing various AI applications such as Nvidia Earth-2, Digital Humans, and robotic factories. In his approximately two-hour speech, Huang discussed accelerating computing trends and how accelerated computing addresses the issue of increased power consumption due to AI.

Trend 1: Accelerated Computing Can Address the Issue of "Compute Inflation"

At the outset, Jensen Huang mentioned that since IBM introduced the 360 central processing unit (CPU) in 1964, the computer industry has a 60-year history. However, with computational demands growing exponentially, the expansion rate of CPUs has significantly slowed down, leading to the issue of "computition inflation."  This term signifies increased electricity consumption in data centers and consequently rising computational costs.

To tackle the issue of computition inflation, NVIDIA has developed graphics processing units (GPUs) and provided CUDA programming tools. These tools enable developers to accelerate various computing tasks, including data processing and deep learning applications, using GPUs. Jensen Huang emphasized that leveraging GPU-accelerated computing can lead to significant cost savings for data centers, both in terms of expenses and energy consumption. This approach aims to achieve a "more you buy, more you save" effect.

Jensen Huang mentioned that "accelerated computing is sustainable computing." By combining GPUs with CPUs, NVIDIA achieves up to 100 times acceleration while only increasing power consumption by 3 times. This results in an energy efficiency of 25 times per kilowatt compared to using CPUs alone.

Trend 2: As AI chip computing power strengthens, energy and time costs decrease accordingly.

Jensen Huang mentioned that aiming to release a new AI chip every year, the latest Blackwell is acclaimed as the "world's most powerful AI chip." It has enabled AI computing power to increase by 1000 times over 8 years. "As GPU computing power increases, its costs also decrease," Huang explained. Using Blackwell to train large language models like GPT-4, the energy consumption required is approximately 1/350th compared to previous requirements, thanks to the increased computational power.

In the past, training large language models (LLMs) with AI could require 1000 GWh of electricity, an amount not even supported by a single GW-level data center globally, with computation times starting from about a month. With the new accelerated computing chip Blackwell, the power requirement has reduced dramatically to just 3 GWh. "It only requires 10,000 GPUs, and the computation can be completed in about 10 days," Jensen Huang stated.

Trend 3: Proliferation of AI Applications in Climate Technology, Digital Humans, and Robotics

In addition, NVIDIA has also implemented the concept of accelerated computing in practical applications, with one of the most ambitious projects being Nvidia Earth-2. Described as creating a digital twin of the Earth, this project directly simulates a duplicate Earth using AI to predict climate changes. It aims to assist countries and businesses in addressing the impacts of extreme weather conditions.

It is reported that Nvidia Earth-2 boasts 12.5 times higher image resolution compared to current numerical models, achieves a speed increase of 1000 times, and enhances energy efficiency by 3000 times. This enables better analysis, planning, and simulation of weather impacts. Currently, Taiwan's Central Weather Bureau under the Ministry of Transportation and Communications utilizes Nvidia Earth-2 to predict more precise landing locations of typhoons.

In addition, in the future, Digital Humans will revolutionize various industries by using generative AI to create digital nurses, customer service representatives, teachers, and more. In simpler terms, this means that "computers in the future can interact as smoothly as humans." For example, an AI interior designer could provide decorating advice and even handle purchasing materials and furniture.

Finally, Jensen Huang pointed out, "The next wave of artificial intelligence is physical AI, understanding the laws of physics, enabling AI to work alongside humans."

Currently, several major Taiwanese electronics manufacturers such as Foxconn, Delta Electronics, Wistron, and Quanta have integrated NVIDIA's autonomous robotics into their factories. They utilize digital twin technology to plan their factories and train robots, aiming to enhance production efficiency and reduce costs. "In the future, all factories will be robot factories, where these robots will manufacture products, including other robots," predicted Jensen Huang.

※ This text is authorized for reprint from "RECCESSARY," with the original title "NVIDIA Huang Renxun Reveals the New Computing Era! What is Compute Inflation? How Do GPUs Solve Power Consumption Challenges?" by Chen Yingxuan. Not subject to CC co-creation licensing terms.

Source:
Environmental Information Center