Take a fresh look at your lifestyle.

Google Tensor is Google’s starting gun in mobile chips: the focus is artificial intelligence, not winning in benchmarks

The new Pixel 6 and Pixel 6 Pro say goodbye to alliances with manufacturers such as Qualcomm: Google has assembled it on their own, and these are their first phones with their own chip called Google Tensor.

The company has already advanced some details about these SoCs (System-on-a-Chip), but now we know more data about some processors that want to give Google more control and an approach that at the moment seems clear to the field of artificial intelligence applied to various scenarios. Computational photography or real-time translation are two good examples of this.

Deep learning by flag

The Google Tensor chip has been developed in collaboration with the Google Research division. The objective, explain those responsible, was create “the best deep learning mobile computer” to apply that power to the smartphones of the Pixel family.

Tensor3

Thanks to these advances they have been able to apply the “magic” of deep learning and artificial intelligence to some of the new camera features of the Pixel 6 / Prosuch as Motion Mode or face blurring.

Five Advantages (and One Disadvantage) Of Google's Tensor Chip: This Looks Great, But A Shadow Haunts Your Future

The design of the chip is focused precisely on providing its power in areas such as voice dictation, natural language recognition, image and video.

We are therefore not facing a conventional rival to other mobile chips: Google has specifically worked in areas that allow, for example, to recognize language – even different dialects and accents. and then translate it in real time accurately.

under
under

In fact, the Google Tensor uses the most advanced Automatic Speech Recognition (ASR) system ever created by Google, and they make use of an ASR model that is capable of working. with applications that work for long periods such as Voice Recorder or real-time subtitles without any noticeable impact on battery life.

Water
Water

But of course those improvements also translate into the field of computational photography and video taking. This is seen for example in the Movement Mode that allows us to go from a photo of a waterfall with static water to a photo in which the water seems to come to life, as if we had taken the photo with a long exposure.

Those improvements also apply to video thanks to HDRNet technology that is now integrated into the Google Tensor chip and that according to Google is capable of working on videos recorded at 4K and 60 FPS to achieve accurate results and more vivid colors.

Finally, these chips also have the Tensor security core and with the integrated Titan M2 chip that allows us to protect our sensitive data and resist attacks with electromagnetic analysis and even laser fault injection.

And what about performance?

Those improvements are interesting, but many will wonder if the performance of the Google Tensor can compete with top-tier chips like the current Snapdragon 888 or Apple’s chips.

Nuclei
Nuclei

That remains to be seen and we will have to check how the new Pixel 6 and Pixel 6 Pro behave in the tests, but what we do know is that the CPU of this SoC is composed of eight cores: two high-performance, two standard-performance, and four high-efficiency cores.

The GPU for its part has 20 cores that will allow you to enjoy “a Premium experience in Android games”, but in Google they did not want to give too many details about the performance that can be expected neither in terms of CPU nor GPU.

Performance
Performance

The only reference they gave was the one that allowed us to compare the performance of these chips with the Snapdragon 765 of the Pixel 5 from last year. Based on their internal testing, the Google Tensor SoC CPU is 80% faster than that chip, while the GPU is 370% more powerful.

At Google they spoke of “heterogeneous computing” in the sense of the versatility of these chips. On the Google Tensor they weren’t looking for peak speeds “that are very good in benchmarks”, but do not reflect the actual use that users make of their devices.

It will be interesting to see how this chip behaves and if it lives up to that Google proposal that seems not to be so surprising when compared to other chips as to offer its advantages in a practical way. For example in these new modes and photographic features of some mobiles that of course seek to regain the throne of mobile photography. We will see if the Google Tensors end up having a lot to say in that challenge.

More information | Google