Preface
In 2023, which has been hailed as the “first year of generative AI”, technology giants have successively launched AI weapons, such as Microsoft Copilot, Bing, Google Gemini, Amazon BedRock…, even Elon Musk announced as the super brain to build Tesla electric cars:Dojo supercomputer, AI has opened up a new game for technology giants.
As the saying goes, during the war, the most profitable people were the arms dealers; during the gold rush, the most profitable people were the shovel sellers.
This is the current NVIDIA Huida existence.
The most powerful AI arms dealer: NVIDIA
I believe you have all seen this picture recently!
NVIDIA's current market value is equivalent to the sum of the eight internationally renowned semiconductor companies on the right！
During the year 2023,NVIDIA stock surges 239% ,
The market capitalization has also taken advantage of this AI craze to overtake Amazon and Alphabet (Google's parent company) and become the third largest company in the United States by market capitalization, second only to Microsoft and Apple.
Why on earth is NVIDIA?
Because NVIDIA’s star product GPU isTraining Generative AI (GenAI) necessary weapons.
Faster computing power and shorter training time make technology companies unable to live without it.
But what exactly is a GPU? Doesn't NVIDIA have any other competitors? What is so special about this company?
Let’s share it with you today NVIDIA Huidas story!
3 Takeaways if you only have 1 minute
1. NVIDIA’s GPU technology:
NVIDIA dominates the market with powerful GPU technology, especiallyDominate the AI and gaming fields.
GeForce series GPU Improved the sophistication of game graphics,
Tensor coreThis will greatly speed up the training and learning of AI models.
And match CUDA platformExpand the application scope of GPU.
2. AI craze:
With the explosion of generative AI, NVIDIA's GPU has become the core tool for training AI models.
Technology giants such as Microsoft, Google and AWS are rushing to use NVIDIA's GPUs to improve AI computing power.
At the same time, there are thousands of AI startups on the market that will rely on NVIDIA's chips for fast and efficient training.The wave of AI has driven a substantial increase in NVIDIA's market value.
3. In addition to GPUs, NVIDIA also has a great business model:
NVIDIA has transformed from a pure GPU supplier toComprehensive systems provider.
By providing complete AI solutions, from the data center technology Spectrum-X required for training, to the training platform Hopper, to the operation platform Blackwell,Selling Total Solutions to Enterprises.
NVIDIA has built a strong platform ecosystem that attracts a large number of developers and enterprise customers.
Taking full advantage of leverage and scale is the secret to NVIDIA's continued competitive advantage.
NVIDIA founding background
NVIDIA has been established for more than 30 years.The following will analyze NVIDIA's growth history in 10-year stages..
Spanning 30 years, from handheld games to outer space Mars simulation, and finally landing on artificial intelligence.
1990s: Focus on the computer game industry
NVIDIA was co-founded in 1993 by Jensen Huang, Chris Malachowsky and Curtis Priem. From the beginning, it focused on making game graphics more refined and beautiful.Graphics Processing Unit (GPU)
(Don’t worry, I’ll explain what a GPU is later!)
In the first year of its establishment, it received US$20 million in investment from venture capital firms such as Sequoia Capital and was successfully listed six years after its establishment.
2000s: In addition to games, we also need to make Mars simulations and automotive chips
During this period, after NVIDIA established its position as the main supplier of gaming chips, it began toExtend your tentacles to the automotive industry and even outer space!
In 2003, NVIDIA became a graphics chip supplier for a number of Audi cars, supporting in-car navigation systems with higher-quality image displays.
At the same time, NVIDIA also cooperates with NASA toGraphics processing technology is used to simulate a realistic Martian environment to help astronauts train on virtual Mars.
2010s: GPUs are perfect for training artificial intelligence!
NVIDIA 的老本行本來是做遊戲的GPU, 但人們漸漸發現，因為GPU 「多核心」的本質，可以同時處理大量數據，所以很適合拿來做需要快速平行運算的事，其中一個就是：訓練AI!
Popular science classroom:
- What is parallel operation?
Imagine having a central kitchen with many chefs responsible for cooking different dishes, rather than one chef doing it from start to finish, so a meal can be completed faster.
The "multi-core" structure of the GPU is like this "multi-chef" kitchen.The ability to process multiple computing tasks in parallel and speed up the overall speed is the concept of parallel computing.
- What is the relationship between parallel computing and AI artificial intelligence?
Parallel operations can process a large amount of data at the same time, and when training AI, a large amount of data calculations are also required to allow AI to learn quickly, so parallel operations can be used to train AI!
CUDA platform
In addition to training artificial intelligence, NVIDIA also developed the CUDA platform (Unified Computing Device Architecture) in 2006.Let the GPU not only do graphics processing, but also do non-graphics calculations.
What is the CUDA platform?
Let’s continue with the kitchen analogy:
GPUs are teams of chefs: chefs who were originally responsible for cooking specific dishes, such as French fries or fried fish.
CUDA is a kitchen management system,Make these chefs more versatile:
CUDA allows these chefs to not only cook the dishes they are good at, but also handle other cooking tasks at the same time.
Such as cutting vegetables, stirring, making soup, etc.
2020s: 30 years of GPU hard work, harvesting fruitful results in artificial intelligence
2022,When OpenAI announced that ChatGPT was the result of training on 10,000 NVIDIA GPUsLater, technology giants such as Microsoft and Google rushed to buy NVIDIA chips to train their AI to be more powerful (investors rushed to buy NVIDIA stocks XD).
This also explains the news at the beginning: In 2023, NVIDIA stock can rise by 239%, and one of the reasons why its market value has overtaken Amazon and Alphabet.
NVIDIA business model
What’s the secret to NVIDIA’s business model?
After learning about NVIDIA's 30 years of great achievements, let's talk about its business model!
At the NVIDIA Investor Relations Conference in April 2016, Jensen Huang mentioned,
NVIDIA's business model is mainly driven by two core elements:
"Platform and Ecosystem" and "Leverage and Scale Effect".
Platform and Ecosystem
The core of this model is to create a platform to attract developers and stick users to the NVIDIA ecosystem.
So how does NVIDIA build an ecosystem?
- Provide additional tools to increase developer stickiness:
NVIDIA improves the quality of developers' games by providing simple tools such as GameWorks SDK (a tool that helps game developers take advantage of NVIDIA's GPUs to make game graphics more refined and realistic). - Expand market scope and attract more customers:
NVIDIA expands its market influence by cooperating with dealers to sell GeForce GPUs (GPUs that can make computer graphics smoother and provide a better gaming experience) to gamers.
- Create a closed-loop ecosystem:
Games developed by developers using GameWorks SDK perform better on GeForce GPUs. The smooth gaming experience attracts more players to buy and more developers to use the NVIDIA platform.Form a virtuous cycle with continuous momentum.
This model has also been copied by NVIDIA to other fields, such as autonomous driving, data centers, etc.
Leverage and scale effects
The essence of this model is to maximize the use of the same technology and apply the same core technology in different markets.
Effective scale and cost sharing can be achieved.
Let’s take a look at how NVIDIA operates the leverage effect:
Multi-market applications:
NVIDIA applies the same GPU architecture to different scenarios.
For example: GeForce focuses on games, Quadro focuses on office, Iray focuses on virtual reality (VR), DRIVE focuses on autonomous driving, and A100 and H100 focus on data centers.
cut costs:
By using the same technology architecture across multiple markets, NVIDIA is able to spread R&D costs and reduce the unit cost of its products.
Different markets have established their own ecosystems:
Each market has its own unique needs and applications. By developing corresponding products for these needs, NVIDIA will gradually form an ecosystem for each market.
These two core business models make NVIDIA like aOperation perpetual motion machine,
It can focus on developing more advanced GPU computing technology, but establishing a good business model and ecosystem will allow NVIDIA to maintain profitability and competitive advantages.
Introduction to NVIDIA’s three major product lines
So is NVIDIA so powerful just by selling GPUs?
As mentioned earlier, NVIDIA's main focus is powerful graphics processing units (GPUs).
These GPUs are used for a wide range of purposes, including gaming, training artificial intelligence, and processing large amounts of data.
Now let us briefly introduce NVIDIA’s mainThree major product lines:
1. GeForce series gaming GPUs
Why are today’s online games becoming more and more realistic? Where does the feeling of being in the scene come from?
NVIDIA's GeForce series of GPUs are for this type of gamecore driving force！
GeForce series GPUs use special graphics processing technology to make game graphics faster and more beautiful.
The light and shadow effects are more realistic. For example, when you are playing a racing game, you can see the reflection of sunlight on the car windows, which is generated with the help of GeForce RTX's Ray Tracing function!
2. AI GPU graphics processor
Why is the current ChatGPT robot so powerful? It can recognize pictures, understand your commands, and has memory and personalized settings at the same time?
of these robotsmain driving forceIt’s NVIDIA’s AI GPU processor H100.
(This type of processor is the GPU used to train AI mentioned in the previous paragraph!)
There are 80 billion transistors in the H100 processor to help AI learn and process data faster.
For example, when you ask ChatGPT to search for information, it can quickly search and compile the information you want, thanks to the powerful support of NVIDIA AI GPU.
3. Arm architecture CPU
First, let’s explain what the ARM architecture is?
The design of ARM architecture is more power-saving and longer-lasting than computer processors, so it is suitable for needsLong-running portable devices:Such as mobile phones, tablets, smart watches, etc.
So what’s so great about NVIDIA’s Arm architecture CPU?
Take NVIDIA's latest processor Grace as an example. Grace is a processor specially designed for artificial intelligence and can improve the computing power and efficiency of computers.
For example: When training an AI model, Grace can process large amounts of data more efficiently, shorten training time, and consume less power during the operation.
This type of CPU can also work together with the GeForce series GPU mentioned above to make the picture run more smoothly. At the same time, Grace CPU can also make supercomputers faster and more power-efficient.Many scientific laboratories use Grace CPU to build supercomputers, which are very versatile!
NVIDIA Customers
NVIDIA funder fathers: Who is buying NVIDIA products?
After reading the product introduction above, it should not be difficult to guess which companies will want to hunt for NVIDIA products!
NVIDIA's main customer groups can be divided into the following three categories:
1. Big tech companies
Customers such as AWS (Amazon Web Service), Meta, Microsoft, Google, etc.
What do these big customers do with NVIDIA's products?
- AWS Use NVIDIA's GPU technology to improve cloud computing performance, make AI training and deployment faster, and allow AWS customers to use more efficient cloud services;
- Meta Use more than 24,000 NVIDIA H100 GPUs to train the next-generation large language model Llama 3;
- Microsoft Use NVIDIA GPUs to accelerate the operation of Azure's AI products. Make it easier for Microsoft customers to develop their own AI on Azure.
Revenue share: FAANG (except Apple) brings NVIDIA 40% revenue!
2. Artificial Intelligence AI Startups
Customers such as OpenAI, DeepMind, Anthropic and Cohere wait.
These companies mainly use NVIDIA GPUs forTraining and application of artificial intelligence models.
For example:
- OpenAI uses NVIDIA GPUs for training ChatGPT for natural language processing;
- DeepMind uses NVIDIA GPUs for training AlphaGo's Go game;
- Anthropic is used to train safe AI systems….
These trainings require a lot of computing power, and NVIDIA's GPUs provide efficient computing suitable for artificial intelligence, improving the speed and efficiency of AI training.
3. Self-driving car companies
Customers such asTeslawait.
For example: Tesla's Autopilot system uses NVIDIA's GPU to process large amounts of data.Quickly analyze traffic data and make driving decisions to ensure safe and accurate autonomous driving.
After reading about NVIDIA’s well-established financial backers, it should not be difficult to understand why NVIDIA’s market value has soared. This is because all technology giants have become red-eyed in this AI war and are desperately buying GPUs to optimize products and retain customers. !
When you see this you may want to ask:
Don’t NVIDIA’s GPUs have any competition?
Is there no GPU from other companies that I can buy?
This company is so strong, there are always some risks to face in the future, right?
What are they going to do next to maintain their current status as leaders?
Because NVIDIA has a long history and a relatively complex product line,
In order to give everyone a relaxing and easy-to-absorb reading experience, we decided to split the article into two parts.
If you have any questions above, remember to lock Hogan & wavelet fans,
We will ride on the Computex craze and continue to provide you with exciting NVIDIA stories.
If you find this article helpful, please remember to continue to lock in the wonderful NVIDIA decryption (Part 2)!
Finally, let’s review the three key points of this article.
3 Takeaways
1. NVIDIA’s GPU technology:
NVIDIA dominates the market with its powerful GPU technology, especially in the fields of AI and gaming.
The GeForce series of GPUs improve the sophistication of game graphics, while the Tensor core greatly speeds up the training and learning of AI models, and is paired with the CUDA platform to expand the scope of GPU applications.
2. AI craze:
With the explosion of generative AI, NVIDIA's GPU has become the core tool for training AI models.
Technology giants such as Microsoft, Google and AWS are rushing to use NVIDIA's GPUs to improve AI computing power.
At the same time, there are thousands of AI startups on the market that will rely on NVIDIA's chips for fast and efficient training. The AI wave has driven a substantial increase in NVIDIA's market value.
3. In addition to GPUs, NVIDIA also has a great business model:
NVIDIA has transformed from a pure GPU supplier to a comprehensive system supplier.
By providing complete AI solutions, from the data center technology Spectrum-X required for training, the training platform Hopper, to the operation platform Blackwell, Total Solutions is sold to enterprises.
NVIDIA has built a strong platform ecosystem that attracts a large number of developers and enterprise customers.
The secret to NVIDIA's ability to continue to maintain its competitive advantage is to draw a complete picture of leverage and scale.
To be continued in the next article…
Related reports
[Silicon Valley Hot Discussion] What is Nvidia’s strength? Who will be the next AI chip giant?
related articles
Taiwan’s first AI unicorn: What is Appier, with a market value of US$1.38 billion, doing?
What is DNS? Introduction to Domain Name System – System Design 06
Introduction to System Design Components Building Block – System Design 05
Back-of-the-envelope Back-of-the-envelope Calculation – System Design 04
Non-functional features of software design – System Design 03
Application of abstraction in system design – System Design 02
English 
繁體中文 
简体中文 
한국어 
日本語 
Español