Google releases self-developed chip Tensor G3 to support enhanced AI functions

[Mobile China News] At the new hardware product launch conference held by Google on the evening of October 4, Google also released a new self-developed chip Tensor G3. This chip has been installed in two mobile phones, Pixel 8 and Pixel 8 Pro. superior.

It is reported that Google has brought a new upgrade to the Tensor G3 chip, paving the way for running generative AI systems on mobile phones. Tensor G3 has an efficient architecture co-designed with Google Research. Google said that the Tensor G3 chip can run more complex ML (machine learning) models, bringing AI enhancements to almost all parts of Pixel 8 and 8 Pro. For example, it can read web pages aloud to users in a "more natural" voice, as well as read them in different languages, and the virtual assistant can also speak in a more natural way.

According to Google, the Tensor G3 chip on Pixel 8 can run twice the number of machine learning models on the device as Pixel 6 (the original Tensor chip). In addition, Pixel 8 Pro claims to be the first mobile phone to run Google's AI model directly on the device, with 150 times the calculation amount of the largest ML model on Pixel 7. In addition, the Tensor G3 chip can also provide better photo and video capabilities, as well as translate face-to-face conversations in 49 languages, transcribe messages in real time, and translate symbols. Google Assistant is also now able to detect and filter out more spam calls.

In terms of performance, the Tensor G3 chip is based on Samsung’s 4nm process and is equipped with 1×Cortex-X3, 4×Cortex-A715, and 4×Cortex-A510 CPU cores. However, according to previously discovered Geekbench 6 running score data, the single-core score of the Google Pixel 8 Pro mobile phone is only 1760, and the multi-core score is only 4442. The performance is not very good.

The above is the detailed content of Google releases self-developed chip Tensor G3 to support enhanced AI functions. For more information, please follow other related articles on the PHP Chinese website!