Foreign media: NVIDIA chip has opened the "crazy" mode
For those who love games, the name NVIDIA must not be unfamiliar. This is a well-known American company that produces high-performance graphics cards that bring a smooth gaming experience to countless players. But you may not know that NVIDIA's chips are not just used in games, but in artificial intelligence.
A few days ago, driven by the large-scale language model of artificial intelligence, the demand for NVIDIA's chips has increased significantly, and the price has also risen, and the market value has exceeded the trillion-dollar mark, becoming the world's first chip company with a market value of more than a trillion dollars. Foreign media directly said that NVIDIA chips have opened the "crazy" mode.
What is an AI big language model?
Artificial Intelligence Big Language Model (LLM) is a large-scale text database based on deep learning, capable of understanding, generating and processing natural language. The model can be used in machine translation, intelligent question answering, text summarization, dialogue systems, content creation, and more.
One of the characteristics of the artificial intelligence big language model is that the number of parameters is very large. As the basic unit of the model, parameters store and update the knowledge of the model. As the parameters increase, the model becomes more complex and can handle more tasks. At present, the parameters of mainstream artificial intelligence language models have reached hundreds of billions, such as Google's SwitchTransformer parameters of 1.6 trillion, Microsoft's ZeRO-Infinity parameters of 1 trillion, and Meta's GPT-3 parameters of 175 billion.
Why are NVIDIA chips so popular?
The training and operation of these artificial intelligence large-language models requires high computing power. The NVIDIA chip provides the key to this function. NVIDIA is mainly a graphics processor, which was originally designed for graphics drawing, but later found that it is also suitable for parallel computing, that is, processing large amounts of data at once. This is the big language model of artificial intelligence.
In order to meet the demand for graphics processors in the field of artificial intelligence, NVIDIA has been continuously introducing a new generation of graphics processors. A few days ago, Nvidia released a new GPU based on the Hopper architecture at the Taipei Computer Show, named H100. The graphics processor has a data throughput of 900 G/s, which is seven times the current industry standard PCIeGen5 interface. NVIDIA also released a generative AI engine called "NVIDIA DGXGH200", which can be used to train and deploy trillions of large AI language models.
NVIDIA not only has strong performance, but also a rich software ecosystem. NVIDIA provides developers and enterprises with a series of tools and platforms for artificial intelligence applications, such as: CUDA, TensorRT, RAPIDS, NeMo, TAO, etc. In addition, NVIDIA has also established cooperative relationships with large cloud computing service providers such as Google, Microsoft, and Amazon to provide users with NVIDIA's chips and services in the cloud computing environment.
What are the challenges of NVIDIA chips?
Despite the popularity of NVIDIA chips, they still face many challenges. The first is because of the insufficient supply of NVIDIA chips, resulting in an increase in chip prices, resulting in some customers being unable to buy. This is because of supply chain issues facing the global chip industry and the limited capacity of Nvidia's foundry, TSMC. Nvidia's chief executive, Jensen Huang, is planning a trip to China to meet with executives from Tencent, ByteDance and others to expand markets and increase production capacity.
Secondly, NVIDIA has also been challenged by competitors in terms of chips, which is undoubted. On the one hand, AMD, Intel and other chip manufacturers are stepping up the development of high-performance graphics processors in an attempt to wrest market share from Nvidia and pocket it. On the other hand, some cloud computing providers and technology giants such as Google, Microsoft, Amazon, Meta are developing their own artificial intelligence chips as a way to reduce dependence on Nvidia. These self-developed AI chips may perform better than NVIDIA's chips in some areas and situations.
In the end, NVIDIA chips also face political, legal and other risks. As the chip dispute between China and the United States intensifies, the U.S. government may impose restrictions or sanctions on NVIDIA, affecting its operations in markets such as China. In addition, NVIDIA is also trying to complete its acquisition of British chip design company ARM, but the deal has been scrutinized by regulators in several countries and has been opposed, so the future is uncertain.
summary
In short, despite the challenges, NVIDIA chips still have broad prospects. As a revolutionary technology, artificial intelligence is changing all aspects of people's lives, creating new value and opportunities. As a leader in the field of artificial intelligence, NVIDIA occupies absolute advantages in technology and market, providing a strong driving force for the development of artificial intelligence.
The CEO of NVIDIA has said that NVIDIA has reached the tipping point of generative artificial intelligence. In other words, NVIDIA's chips will have a broader market and a broader customer base. Therefore, foreign media will directly say that NVIDIA chips have opened the "crazy" mode.