Software development joke

If NVIDIA, a supplier of graphics chips, AI chips and data center chips, is a software company, it may be ridiculed by many people as "unprofessional", but in fact, NVIDIA is becoming a chip company that relies on software platform drivers to make money. According to insiders, 75% of employees in NVIDIA are R&D personnel, and half of R&D personnel are software developers, which is enough to witness the position of software platform strategy in its company. Invista has always claimed that it is a "computing platform" company, not a chip company.

When Yingwei was established, there were more than 20 graphic chip companies in the market, and there were more than 70 graphic chip companies in the market three years later. NVIDIA can ride the dust without the CUDA programming model and data center GPU platform developed by NVIDIA 15 years ago, which brought NVIDIA to the general computing chip market.

With customized software platforms, development tools and application tools, NVIDIA has built a "strong, robust and self-sufficient ecosystem composed of software, universities, start-ups and partners" (Forbes evaluation).

By developing a specific software platform, NVIDIA has continuously expanded its territory, surpassing Intel, AMD, Samsung and TSMC to become the world's highest semiconductor company by market value.

At the GTC conference held in NVIDIA a few days ago, Huang Renxun, the founder and CEO of NVIDIA, who was dressed in leather and once again "talking in Kan Kan" in his kitchen, did not talk about hard chips, but listed several new software platforms, each of which may create tens of billions of dollars of new market opportunities for NVIDIA.

Huang Renxun can always touch the nerves of industry and capital, because his chip strategy is always to dream of opening the way with the platform. When the developer comes to his specific platform as scheduled, it will naturally bring about a big sale of chips. Its dream and platform priority strategy have never failed, and now it is showing a snowball effect. At this GTC conference, Huang Renxun revealed that there are nearly 3 million developers in NVIDIA. In the past 15, the download volume of CUDA was 30 million times, and it reached 7 million times last year.

Now, NVIDIA once again copied the platform strategy to the meta-universe market. This year, the Metauniverse is in full swing, and NVIDIA started the research and development of its Metauniverse platform as early as two years ago. June 5438+February last year, NVIDIA's cloud universe platform Omniverse was officially launched for public beta. Huang Renxun doesn't brag about the cosmos platform like Facebook, but calls it a real-time simulation and collaboration platform for more than 40 million engineers and designers around the world. However, this collaboration platform is highly respected by all walks of life.

Recently, NVIDIA surpassed Berkshire Hathaway to become the seventh largest company in the United States. Wells Fargo raised NVIDIA's share price from US$ 245 to US$ 320 on the grounds that "NVIDIA will be the hardware entrance to the meta-universe". In the next five years, Metauniverse will provide NVIDIA with a market share of 654.38+000 billion US dollars. And this engine that can bring an increase of $654.38+0 billion is Omniverse.

Mario Stefanidis, vice president of research at Roundhill, believes that NVIDIA's Omniverse platform is compatible with virtual reality, allowing users to collaborate in real time and create accurate physical simulations with real light. "This type of infrastructure will be the key to the virtual world, because users will seek an immersive experience and connect the real world with the virtual world."

Regarding NVIDIA's metauniverse strategy, Wang Congqing, president of HTC China, believes that NVIDIA, as an infrastructure technology provider providing PC/server/ cloud graphics solutions and content creation tool solutions, has many cloud-based XR solutions/tools for transferring work processing to the network; NVIDIA cooperates with many participants in the industry and tries to realize their plans. It should be said that NVIDIA's meta-cosmic strategy will bring high value to the industry. "When a company makes it clear that they play a role in the whole metauniverse story, instead of being the master of the whole metauniverse, they will have a greater chance of success." Wang Congqing said.

Just a few days ago, Time magazine awarded the 20021Best Invention Award to Omniverse. The evaluation is that "the virtual world is not only built for games-it is also helpful for the planning of infrastructure such as roads and buildings, and can be used to test self-driving cars. Extremely realistic virtual space helps to achieve real-world goals, and the Omniverse project in NVIDIA is making it easier to create it. The platform combines the latest ray tracing technology of NVIDIA GPU with a series of open source tools, allowing real-time collaboration in a realistic 3D world. Architectural experts such as Foster Architects are using this technology to visualize architectural details in the early design stage. BMW used it to build digital twins of a factory to test the possibility of a more efficient assembly line.

Meta (Facebook) Meta Universe's target customers are after-dinner entertainment and gamers, while NVIDIA is a corporate engineer and designer. Metauniverse may be icing on the cake for mass consumers, but it is "necessary" for enterprise users and entrepreneurs to create productivity. In fact, NVIDIA's competitors are not Meta (Facebook), but AMD and Intel. 1October 9th, GTC 165438 was held in NVIDIA. AMD and Facebook announced a cooperation, and Facebook will use AMD's data center chip. This news made AMD's share price hit a record high. As a chip manufacturer, AMD needs a Facebook platform to get in touch with the upper metauniverse application ecology, and NVIDIA has its own platform.

Omniverse started online testing at the end of last year, and now it has more than 700 enterprise users, including BMW, Ericsson, aerospace manufacturer Lockheed Martin and Sony Animation.

NVIDIA introduced that Omniverse is divided into three parts. The first part is the database engine Omniverse Nucleus, in which users can connect and exchange 3D assets and scene descriptions, and designers who need modeling, layout, shadow, animation, lighting, special effects or rendering can create scenes together. The second part is the synthesis, rendering and animation engine-the simulation of virtual world. For example, NVIDIA's graphics technology can simulate how every light in the virtual world is reflected in real time. The third part is NVIDIA CloudXR, including client and server software, which can be used to transfer augmented reality content from OpenVR applications to Android and Windows devices, allowing users to enter and exit Omniverse.

At this year's GTC, NVIDIA released a series of performance updates and latest tools of Omniverse, and held a special session on Omniverse technology. Huang Renxun said, "With Omniverse, we have the technology to create a new 3D world or simulate the physical world. How to use Omniverse to simulate digital twins in warehouses, factories, physical and biological systems, 5G edges, robots, self-driving cars and even virtual images is an eternal theme. "

NVIDIA's software platform, development tools and application development tools have won a lot of "development powder", of course, not because of Huang Can's bluff. Since the PC era, NVIDIA has attached importance to tools and been considerate to developers. Now this tradition is carried forward in the meta-universe.

One is Omniverse Avatar, which is a development tool for creating virtual people. Based on the Omniverse platform and Megatron 530B, the world's largest customized language model, this tool allows developers to conveniently develop interactive virtual images driven by AI, so that this character can have visual and language skills, talk about various topics and understand the deep meaning in natural language.

In order to let developers see the ability of NVIDIA's Avatar, Huang Renxun showed the avatar "Toy-Me" of Huang Renxun, and also brought the virtual human "Eggshell Man" who helped people order food and the virtual human who could be embedded in conference software.

The second is Omniverse Replicator, which is a synthetic data generation engine that helps to build better digital twins. Replicator applications enable developers to guide AI models, fill data gaps in the real world, and mark real data in ways that humans cannot.

In GTC's keynote speech, NVIDIA also demonstrated a series of new functions of Omniverse, including exhibition hall, farm and Omniverse AR. Omniverse VR will be the first to launch a VR environment with full image and real-time ray tracing.

As a platform semiconductor company with a rising global market value, it will certainly not stick to the meta-universe. In the fields including autonomous driving, robotics, global climate modeling, large-scale models, data centers and so on. No one in NVIDIA can be left behind, all of them have been laid out, and relevant platforms, frameworks, tools and algorithm models have been launched.

Autopilot opens a trillion-dollar market. Huang Renxun believes that "by 2024, most new electric vehicles will have real automatic driving function." For autonomous driving, NVIDIA can't be absent, so he launched NVIDIA DRIVE, which is a full-stack open platform for autonomous vehicles in NVIDIA, and Hyperion 8 is the latest complete software and hardware architecture in NVIDIA. At this conference, Huang Renxun introduced several new technologies built in Hyperion in detail, including Omniverse Replicator for DRIVE Sim. In fact, many leading enterprises and entrepreneurial teams in the field of autonomous driving have adopted NVIDIA's end-to-end autonomous driving solutions, including Lotus, Baidu, Weimar Automobile, Gaohe Automobile, Weilai, Light and Heavy Zhixing and so on. Even Tesla adopted NVIDIA's GPU and system platform in its early days.

Robots are a battleground for military strategists, and Intel is also heavily deployed here. At present, NVIDIA's robot ecosystem VIDIA Isaac has more than 700 companies and partners, which has increased fivefold in the past four years. Huang Renxun announced at GTC that the NVIDIA Isaac robot platform can now be easily integrated into the robot operating system (ROS). This is a set of widely used software libraries and tools, suitable for robot applications. Huang Renxun explained that Isaac Sim built on Omniverse is the most realistic robot simulator ever.

The big model is regarded as a possible route for general AI and a strategic location for AI. At present, IT head enterprises are all laid out here. For large-scale models, Huang Renxun released Nemo Megatron for training large-scale language models. He said that this model "will become the largest mainstream HPC application in history." At the same time, NVIDIA also released NVIDIA Modulus to build and train physical machine learning models, which can learn and fully abide by physical laws.

In addition, Huang Renxun also announced the launch of a new Clare Holoscan platform to create better medical and health services; The launch of the quantum computing simulation platform has significantly improved the performance of the quantum model. It will be launched in June+February, 5438, and Earth II (E-2) will be built to simulate and predict climate change. "In order to create Earth II, we need to use all the technologies invented so far." Huang Renxun said.

However, regarding whether NVIDIA is a software company, Wang Congqing thinks: "Not really, NVIDIA is a tool and infrastructure provider. Instead of making content, they provide some underlying technologies/tools for others to make. " At the beginning of GTC's speech, Huang Renxun said that he was AI: "NVIDIA, deep learning and ubiquitous human wisdom have integrated me into everyone's life."

In the era of digital economy, software is defining everything. We use software to build everything. How can we leave the chip? This is the real logic for NVIDIA to build computing platforms, development tools and application tools.

Author Li Jiashi

Editor Lian Xiaodong

Bian Mei Maria