Google announced its new Tensor G5 processor |
|
![]() |
|
Along with the Pixel 10 series of smartphones, Google officially announced the Tensor G5 processor. This is the company's first chip based on TSMC's 3nm process technology, rather than Samsung's, which in itself is a significant change. The manufacturer talked about all the advantages and features of the new platform. According to the developers, the new product provides an average of 34% higher performance compared to last year's Tensor G4, as well as an increase in energy efficiency, allowing smartphones to work for several hours longer. The architecture of the Tensor G5 has changed significantly: it now uses a ?1 + 5 + 2? scheme instead of ?1 + 3 + 4,? and the maximum frequency of the senior core reaches 3.78 GHz. Google has separately noted the acceleration of AI tasks ? the built-in TPU has become 60% more powerful, opening up new possibilities for processing data directly on the device. For users, this means faster performance of Gemini Nano ? the built-in AI model, which is now 2.6 times faster and twice as efficient compared to the previous generation. An interesting innovation was the introduction of the proprietary Matryoshka architecture for working with transformer models: Tensor G5 is capable of processing up to 32,000 tokens at a time, which is roughly equivalent to a hundred screenshots. This significantly expands the boundaries of local generative processing of text, images, and audio without an internet connection. The video component has also been improved: the updated ISP reduces blurring when shooting in low light conditions and makes 10-bit HDR ![]() However, the chip does have limitations. The Tensor G5 lacks hardware ray tracing support, which leaves mobile gaming as a strong point for competitors such as the Snapdragon 8 Elite or Dimensity 9400+. |