Google Tensor, the CPU of the Pixel 6 and 6 Pro promises to be very particular

The advances in the Pixel 6 / 6Pro house continue: after yesterday’s rumors, the colleagues of xda-developers continued to dig in collaboration with their anonymous source, focusing on the proprietary Google Tensor SoC, and have discovered some really curious details. Meanwhile, both models were exhibited at the Google Store in New York : see some images shared by users on social networks around the article. In Japan, on the other hand, Google has published a nice post talking about its chip, playing on the double meaning of the word in English, which is also used for potato chips in bags. We include it below.

Like many Linux based / derived operating systems, in the Android filesystem there is a file called / proc / cpuinfo . It is a plain text document, which is populated by the processor itself and which is read by the kernel; it contains, as the name implies, information on the CPU, and in case of multi-core configuration the detailed technical specifications of each of them. By reading this file, it was possible to rebuild the Tensor CPU hierarchy, and it is something that has never been seen before :

  • 2 Cores Cortex-X1 to 2, 802 GHz
  • 2 Cores Cortex-A 76 to 2, 253 GHz
  • 4 Cores Cortex-A 55 to 1, 80 GHz

The Cortex-X1 were presented by Arm in the 2020, and are the most powerful of all. They are found on top of the range chips such as the Snapdragon 888 and the Exynos 2100, but no chip has so far implemented them two. This is already a remarkable oddity, but the use of the Cortex-A 76 as intermediate cores it is even more so: it is a very, very old design , presented by Arm even in 2018 (found for example in Snapdragon 855). For comparison, the S 888 and the Exynos 2100 depart from Cortex-A 78, presented in 2020 together with the X1, which is clearly superior in terms of power, consumption and size. As the source says, there doesn’t seem to be a rationale for Google’s choice, at least with the information we have so far.



It should be noted that the Cortex-A 55 are old (presented in 2017, even), but the speech is very different, because there have been no concrete successors until to this spring with the introduction of the ARMv9 architecture that changes all the cards on the table a little. It generally takes a long time from introducing Arm architectures to implementing them in a processor (typically: architectures in spring, first SoCs at the end of the year / beginning of the next), and it is actually unrealistic to expect them on a chip that will be released after only 4 months. All this to say that if you are looking for low-power cores, the Cortex-A 55 are still the current option, while for A 76 there are even two more recent generations (always leaving out ARMv9).

As we said, the information is been extrapolated from a system file managed by the processor itself; in short, it is a reliable source, with very low chances that it has been tampered with . Nothing is safe until officially, but it appears that Google and Samsung have actually assembled a very peculiar “Frankenstein” CPU. A rather interesting theory is that it was decided to use two Cortex-X1s precisely to compensate for the A’s 76, which at this point could be a key component to the functioning of other parts of the chip (the number one suspect being the TPU, of course ). But these are just conjectures.

A few days ago the SoC appeared on GeekBench, and showed results rather low – not top of the range, we said, at least not current. Even with what we have discovered, it is our duty not to worry too much , because we do not know which core was used for the single-core part of the test, and in any case the maximum frequency was significantly lower than the peak frequency – 2, 15 GHz. We know that Samsung likes to benchmark its processors in power saving mode (a example of this behavior we saw just a few days ago with S 22 Plus).


Back to top button