Last month, Nvidia became the first chip manufacturer with a market value of $2 trillion. In addition to focusing on providing hardware support for artificial intelligence models (such as H100 and the upcoming H200 GPU), the company's software and hardware collaborative system CUDA computing platform is also a major cash cow, which has created Nvidia's dominant position.
However, according to reports, some large technology companies are teaming up to try to break Nvidia's current dominant software advantage in the artificial intelligence market - the CUDA platform.
CUDA architecture
Nvidia's CUDA computing platform has enabled Nvidia to achieve its current peak in market value. It is both a parallel computing platform and a programming model, containing a series of development tools. CUDA's extremely high technological maturity, absolute performance advantages, and extensive ecosystem support make it the most commonly used and popular collaborative platform in artificial intelligence research and commercial deployment.
The CUDA architecture enables developers or software engineers to better utilize the powerful computing power of Nvidia GPUs and accelerate computationally intensive tasks through software hardware collaboration, not limited to traditional graphics rendering, but also including deep learning, scientific computing, image processing, and more.
However, CUDA only supports NVIDIA GPUs and is not compatible with other mainstream GPUs such as AMD and Intel.
In his keynote speech at the 2023 International Computer Show, NVIDIA CEO Huang Renxun stated that 4 million developers are using the CUDA computing platform.
Therefore, in the eyes of some industry insiders, Nvidia's ability to gain a monopolistic share in the artificial intelligence chip market is not only due to its high-performance GPU chips, but also due to its indispensable CUDA architecture. The combination of the two is superior to any chip produced by other chip manufacturers currently.
Joint attack
However, the explosive demand for Nvidia products in the market has led to shortages, which has also given some competitors the opportunity to continue developing related products.
According to reports, an organization composed of Intel, Google, Arm, Qualcomm, Samsung, and other technology companies is developing an open-source software ecosystem to prevent artificial intelligence developers from being "trapped" in Nvidia's proprietary technology.
This joint project, named UXL, aims to create an open-source artificial intelligence hardware and software ecosystem that will allow developers' code to run on any hardware and chip. The technical specifications of the project will be determined in the first half of this year, and its technical details are expected to reach a "mature" state in the second half of this year, but UXL has not yet given a final release target.
It is reported that the project will include the OneAPI open standard developed by Intel, which aims to eliminate barriers to specific coding languages, libraries, and other tools that bind developers to specific architectures.
"We are actually showing developers how to migrate from the Nvidia platform and provide a diverse computing platform," said Vinesh Sukumar, head of Qualcomm's artificial intelligence and machine learning, in an interview
"This is about how we can create an open ecosystem that promotes productivity and hardware choices in the context of machine learning frameworks," said Bill Magro, Google's Director of High Performance Computing and Chief Technology Expert
UXL stated that although the initial goal of the project was to provide more choices for artificial intelligence applications and high-performance computing applications, and win over a large number of developers for its platform, in the long run, UXL's ultimate goal is to support Nvidia's hardware and code.
When asked about the efforts of open source and venture capital software to break Nvidia's dominant position in the field of artificial intelligence, Nvidia executive Ian Buck responded in a statement, "The world is accelerating. The new ideas for accelerated computing come from the entire ecosystem, which will help advance the scope that artificial intelligence and accelerated computing can achieve."
Jay Goldberg, CEO of D2D Advisory, a financial and strategic consulting firm, pointed out that Nvidia's software and hardware position in the AI field is currently difficult to shake. "What's important is that people have been using CUDA for 15 years, writing code and optimizing work around it."