A ONE Institute
Nov 2, 2024
Today, I will share a post about industry trends, focusing on NVIDIA.
Understanding current industry trends is crucial when considering students' future prospects. This post is aimed at briefly summarizing recent industry trends, starting with a look at NVIDIA, the company making waves worldwide. Reports have been highlighting NVIDIA as a company valued at $3 trillion and a leader in the AI market.
Let's begin by exploring NVIDIA’s founding journey. A computer's main components include the CPU, RAM, and Hard Disk.
CPU (Central Processing Unit): Equivalent to the human brain, responsible for processing data.
RAM: Like a desk where open books can be referenced, RAM is a temporary memory storage that holds data only while the power is on.
Hard Disk: A long-term storage device where data can be saved and retrieved even after the power is off.
If you look inside a computer, as shown in the above image, you can see that the CPU handles complex computations overall. Data is spread out in RAM for processing, and, for long-term storage, it moves to the Hard Drive.
In the early 1900s, Intel dominated the market by producing CPUs, playing a central role as the producer of the core component that functions like the human brain, making it the leading company at the time. To keep up with advancing computers, they worked on CPUs capable of better visual effects and audio processing. However, as data to be processed increased, the time for computation lengthened, creating bottlenecks in computers where speed was paramount.
At that time, Jensen Huang, a Taiwanese engineering student, observed the limitations of CPUs when used for general purposes. Inspired by this, he aimed to create a chip specialized in rendering graphics and founded NVIDIA. The name "Invidia" comes from the Latin word for envy, which he adapted to "NVIDIA."
NVIDIA's specialty was developing chips that maximized graphic performance, creating what is known as the GPU (Graphics Processing Unit). Unlike the CPU (Central Processing Unit), which has multiple cores, a GPU is characterized by having thousands of cores. While a CPU performs serial operations, processing one calculation at a time, a GPU performs parallel operations, handling many calculations simultaneously and sending results to memory.
To illustrate, consider rotating a sketch of a car by about 10 degrees using three axes. The CPU processes this by calculating matrix operations involving calculus and trigonometric functions like sine and cosine. In contrast, the GPU, designed for graphics, divides the task among its numerous cores for parallel computation.
Processing each point sequentially can be fast on a CPU but becomes inefficient when many simple operations are needed. In such cases, the GPU’s speed excels as it performs many parallel calculations.
NVIDIA was the first to develop the GPU, an essential chip for video processing and computer gaming. By consistently releasing new products, NVIDIA dominated the GPU market and experienced remarkable growth. The above image shows how graphics evolved with the development of these chips—from looking like cartoons to movies and eventually appearing realistic.
However, the company faced challenges during the dot-com bubble in the 2000s and the financial crisis of 2008, nearing bankruptcy at one point. Jensen Huang even took a $1 salary to focus on saving the company. Success sometimes requires a bit of luck, doesn’t it?
Many of you might remember AlphaGo, an AI developed through deep learning. Deep learning requires significant parallel computation involving vectors and matrices, where the GPU shines. Training AI on GPUs was found to be efficient and time-saving. Additionally, cryptocurrency mining, which became prominent in 2017, also benefited from GPUs, shortening processing time.
Although NVIDIA initially developed GPUs with graphics in mind, they turned out to be ideal for AI deep learning and cryptocurrency mining. The latest breakthrough has been in LLMs (Large Language Models) and Generative AI. Among various AI applications, including ChatGPT, LLMs are trained using deep learning, which led to a surge in interest in NVIDIA’s GPUs as the ideal tool for training these models.
Since the early 2010s, Jensen Huang and the company recognized this and began producing GPUs optimized for cryptocurrency mining, introducing GPGPU (General-Purpose Computing on GPU).
Hardware development: Transforming GPUs to serve purposes beyond graphics.
Software development: Creating CUDA, a software for efficient GPU commands that enhances GPU performance. CUDA is a programming language that helps GPUs process commands efficiently and has been made freely available to various companies, much like Tesla’s open-source initiative to expand its ecosystem and establish dominance in the market.
Numerous companies working in AI now use NVIDIA’s GPUs and the CUDA software for optimization. Although many strong competitors have emerged with similar products, NVIDIA’s widespread use of GPUs and CUDA has made switching to other companies difficult.
To maintain their competitive edge, NVIDIA unveiled BLACKWELL at the GTC 2024Â conference. The B100Â product release was followed by a combination of two GPUs to create B200. Two B200 units, paired with a CPU, formed GB200, and by linking 36 GB200 units, they developed NVL72, which boasts 30 times the efficiency of its predecessor, the H100.
While most companies see a 10-20% performance improvement between product generations as innovation, NVIDIA has achieved a 30-fold leap. This demonstrates NVIDIA’s technological superiority. Although I couldn’t predict everything precisely, I correctly anticipated in my 2021 presentation that NVIDIA would become a top 3 company. NVIDIA now holds a 98% market share in data center GPUs, with its stock price soaring as a result.
The 2021 presentation aimed to highlight industry trends, and despite some changes by 2024, NVIDIA, ranked 8th in 2021, has risen to 3rd place in 2024. Although other companies continue to develop high-performance chips and challenge NVIDIA’s dominance, the industry remains largely dependent on GPUs and CUDA, which NVIDIA has established as industry standards.
As industries become more technologically intensive, CEOs who lack a solid understanding of technology may struggle to make accurate assessments. Concluding today’s post, I note that many leading company CEOs have engineering backgrounds.
I plan to create a series of videos on this topic and appreciate your interest.
Today, we explored industry trends. I hope this has been helpful in planning for children’s futures. Thank you.