According to foreign media reports, at present, Internet giants such as Google, Facebook, Microsoft, Amazon and Baidu are exploring a variety of chip technologies that can promote the development of artificial intelligence. Their choice will change the fate of chip makers such as Intel and Nvidia. But at present, even the computer scientists of these Internet giants are not clear about the future direction.
Internet giants rely on data centers with thousands of servers to provide services. Each server's engine is a CPU (central processing unit). However, with the use of an artificial intelligence technology called deep neural networks, these companies began to adopt other types of processors than CPUs. By analyzing massive amounts of data, neural networks can learn to perform tasks ranging from recognizing faces and objects in photographs to language translations, which require other types of processors than CPUs.
Therefore, Google developed the Tensor Processing Unit (hereinafter referred to as "TPU"). Microsoft uses a processor called Field Programmable Gate Array (hereinafter referred to as "FPGA"), while other companies use a graphics processor (hereinafter referred to as "GPU"). . They are all considering new chips that accelerate the artificial intelligence tasks of smartphones and other devices.
The choice of these companies is important because their network business is very large. They purchase and operate more computer hardware than any other company, and as the importance of cloud computing continues to increase, their lead will continue to expand. If Google chooses a processor instead of another processor, it could fundamentally change the chip industry landscape.
TPU poses a threat to companies such as Intel and Nvidia because Google produces the chip itself. But GPUs also play a very important role in Google and its fellow brothers, and NVIDIA is the maker of these GPU chips. Intel also joined the camp by acquiring Altera. The $16.7 billion acquisition of Altera is the largest acquisition in the history of Intel and reflects the dramatic changes in the chip market.
However, it is quite difficult to use a type of processor to meet the needs of the Internet giant because the operation of the neural network is divided into two phases. The first phase is the training phase. Companies such as Google train neural networks to perform a specific task, such as identifying faces or translation languages ​​in photos; the second phase is the execution phase, which is the stage in which ordinary users use neural networks, such as The user posts a photo of the high school class reunion on Facebook, which automatically labels everyone in the photo. These two phases are very different and have different requirements for the processor.
Currently, the GPU is the best option during the training phase. Chip makers design GPUs to render graphics for games and other applications that require high graphics processing power, but in recent years, companies such as Google have found that such chips offer an efficient way to train neural networks, which means they Train more neural networks with less hardware. Microsoft artificial intelligence researcher XD Huang called the GPU "a real weapon." Recently, his team completed a system that recognized the language like a human being, and they trained the system for about a year. If you don't use a GPU, it takes 5 years to train this system.
But companies also need chips that can execute neural networks quickly—a process called reasoning. Google has developed TPU specifically for this purpose. Microsoft uses FPGAs and Baidu uses GPUs. These chips are not as good in reasoning as they are in training, but they can also perform inferential tasks with the right software.
At the same time, other companies are developing chips that are suitable for implementing neural networks on smartphones and other devices. IBM is developing such a chip, although some insiders have questioned the validity of such chips. Intel and Movidius reached an acquisition agreement, which has already used such chips in their devices.
Intel is aware of the changes in the market. Four years ago, Intel said it sold more server chips to Google than the other four companies, so it understood the impact of Google and its fellow brothers on the market. Intel acquired technology, in addition to Altera and Movidius, it also acquired a third chip company called Nervana.
It makes sense for Intel to do so because the market is just beginning to develop. Intel vice president Jason Waxman recently said, "We are currently on the verge of the next wave of rapid growth, and the next wave of high-speed growth is artificial intelligence."
Spot Lights Moving Head ,Led Moving Head Spot,Moving Head Spotlight,Spot Moving Head Light
Guangzhou Cheng Wen Photoelectric Technology Co., Ltd. , https://www.cwledwall.com