In addition to publishingThe launch schedule for the large-scale natural language model "Gemini"At the same time, Google also unveiled the next-generation TPU v5p, which boasts scalability and flexible deployment features. It is also Google's most powerful tensor accelerator to date.
"Gemini" is throughTPU 4AndTPU v5eThe two TensorFlow accelerators are used for training, and the actual deployment and operation is through the newly announced TPU v5p, which is claimed to be able to achieve twice the computing power on a single processor and can correspond to the next-generation artificial intelligence models with four times the scale. It can also be used to train existing models, reducing the training time by 2 times.
In addition to being used to promote artificial intelligence computing, Google also emphasized that the new generation of TPUs can accelerate the operation of more cloud tasks, while also further improving the execution efficiency of Google's online services such as search, YouTube, Gmail, Google Maps, and the Google Play Store. It can also enable many Android devices to integrate cloud-based collaborative computing application services to have faster execution efficiency, and combined with device-side computing, provide users with a more convenient user experience.
In addition, Google also emphasized that the new generation TPU v5p not only provides higher computing performance but also reduces overall power consumption. It can also be compatible with next-generation artificial intelligence models through its own scalability and meet the needs of developers and enterprises investing in large-scale automatic generative artificial intelligence computing training.
By announcing the launch of the next-generation TPU v5p, which will serve as the basis for "Gemini" computing acceleration, Google is clearly using this opportunity to illustrate that its development of artificial intelligence applications will not be limited by its competitors' reliance on partners to provide GPUs, and emphasizes that it can promote the growth of its artificial intelligence technology in a self-sufficient manner.
