Google says its Tensor chip is 80% faster than the Pixel 5’s CPU
Google has been talking about its new ‘Tensor’ System-on-a-Chip for a while now, but we’ve been in the dark regarding specific details. Today is the launch day for the Pixel 6, the first device powered by Google Tensor, and Google has finally revealed technical details about the chip.
Google is calling Tensor the best hardware innovation in the company’s history, and the name is a nod to TensorFlow, the machine learning library that Google has offered for several years. Tensor has four ARM Cortext A55 cores, which are slower and more power-optimized, paired with two 2.25GHz A76 cores, and two Cortex X1 cores running at 2.8GHz for foreground processing (according to Ars Technica).
Google says Tensor is 80% faster than the CPU on the Pixel 5, which used a Snapdragon 765 5G chipset. The integrated GPU is supposedly 370% faster than the Pixel 5’s graphics. Google told Ars Technica that this design is most efficient with “medium” workloads, like having the camera app open or browsing a web page. We’ll have to wait for real-world testing to know for sure if the unusual chipset layout pays off, but at least from the numbers Google has provided, Tensor seems impressive.
Tensor also has upgraded (and integrated) versions of the independent chips found in previous Pixel 6 phones. There’s an ML Engine component that handles machine learning and AI workloads, specifically optimized for Google’s machine learning models, as well as an “ultra-low power engine” for features like Now Playing music detection.
There are still some performance details that Google isn’t discussing yet, such as benchmarks for AI-optimized tasks, which is one of the main workloads that Tensor is designed for — it’s even in the name. However, once orders for the Pixel 6 start shipping, we should have a better idea of where Tensor lands in comparison to Qualcomm and Apple’s smartphone chipsets.