Previously, the TPU was used for internal Google products, with the opening of the Cloud TPU, which means that AI chips and public cloud market will usher in a new situation.
Google also released a beta version of Cloud TPU, available via Google Cloud Platform, to help more machine learning specialists run models faster. The Cloud TPU is a Google-designed hardware accelerator designed to accelerate and scale specific TensorFlow machines Optimized for studying workloads.
'Open up the TPU for your own use as a cloud service and show that Google is investing heavily in cloud operations and that Google will compete with all vendors that offer smart cloud services and the chip will compete with NVIDIA.' 'Gartner Research Vice President Shengling sea on the first financial reporter said.
Nonetheless, both Google and NVIDIA have denied that there is a direct competitive relationship between the two sides because NVIDIA's chips are geared toward broader applications, while Google's AI chip focuses solely on machine learning, buoyed by the need for artificial intelligence, NVIDIA's shares rose more than six times in the past two years.
As early as May 17 last year, Google I / O General Assembly, Google announced the official launch of the second generation of artificial intelligence focused on the TPU processor. TPU is a self-developed Google-designed for artificial intelligence computing services, high-performance processing For example, the first generation of products have been used in data centers and artificial intelligence including AlphaGo, the main computing performance.
The second-generation TPU chip, released last year, not only deepens the AI's ability to learn and reason, but Google is serious about bringing it to the market.According to Google's internal testing, the second-generation chipset for machine learning training Speed than the current market graphics chip (GPU) save half the time.
Li Fei-fei, chief scientist of Google's cloud computing team and director of AI Labs at Stanford University, said at the time: "These TPUs provide staggering 128 trillion floating-point operations and are chips designed to drive machine learning technologies." In contrast, The iPhone 6 delivers 100 trillion floating-point operations and the NVIDIA Volta GPU delivers 120 trillion floating-point operations.
NVIDIA is currently the dominant GPU GPU, said NVIDIA NVIDIA GPU Cloud (NGC), an advanced neural network that is gaining momentum for the rapidly evolving AI community for rapidly growing global audiences. " NGC has provided services to Amazon's cloud platform.
Google launched the second generation of TPU, it launched Cloud TPU cloud computing services, users can rent, billing and GPU computing services will also take the same mode, according to the minute pricing.At that time, Google said that this latest AI Chips will not be sold to data centers based on companies on other chip maker servers, such as Dell.
This also means that Google's AI chip development ideas not to compete directly with the GPU, but the use of TPU differentiated development in the public cloud industry.Google's chief technology officer UrsHolzle said: 'In essence, TPU is one for Machine-learning supercomputer.Our goal is to provide the best cloud services. '
Google announced the latest quarterly earnings in early February, meanwhile, it also announced for the first time the revenues of its single-quarter cloud business to reach 1 billion U.S. dollars compared with a single-season revenue of more than 5 billion U.S. dollars in both Amazon and Microsoft Cloud, The service market is currently ranked third.
From the perspective of NVIDIA, according to the latest quarterly earnings report, NVIDIA data center business is currently the second largest source of revenue for the company, more than 600 million US dollars, up more than doubled, an increase of 105% last September in China GPU technology conference, NVIDIA announced a partnership agreement with China's BAT for a data center.
Google's entry worries NVIDIA investors who believe Google's involvement will wipe out large data center operator customers, which will affect NVIDIA's earnings and revenue because even Google can not commercialize its chips, Win the bargain.