The second era SoC Tensor G2 appeared with the Pixel 7 and Pixel 7 Star cell phones: what it is, how it works, and its main highlights. Toward the beginning of October 2022, Google disclosed its new Pixel 7 cell phone. Organization spokespersons brought up that the Pixel 7 and Pixel 7 Star utilize the Google Tensor G2 SoC ( Framework on-a-Chip ), where “G2” means “second era.” Yet what are the qualities of the chip that oversees the activity of the new cell phones?
Google has acclimated us to the arrival of its new cell phones a couple of months after the send-off of the highest point of the scope of other Android gadget makers. The first generation of Google Tensor SoC that equips last year’s Pixel 6 (presented in October 2021) hosts eight cores divided into three clusters: the two most powerful cores are ARM Cortex-X1 at 2.8 GHz.
Other manufacturers, such as Samsung for its Galaxy S21, had previously used the same Cortex-X1 cores in their flagship smartphones. Regarding SoC manufacturers, Cortex-X1 was previously used in the Samsung Exynos 2100 and Qualcomm Snapdragon 888. The difference, however, is that its SoC Tensor has inserted two Cortex-X1 cores instead of just one, as did the competition.
A few months after the Google Pixel 6, ARM presented the new Cortex-X2 that Samsung immediately used in its Exynos 2200 SoC of February 2022. With the launch of the new Pixel 7 in October 2022, we expected a Google SoC with two Cortex-X2 cores. The Mountain View company, on the other hand, has proposed smartphones that mount a new generation Tensor G2 SoC but it once again contains two high-performance Cortex-X1 cores, like the predecessor.
The maximum clock frequency declared is 2.85 GHz, practically identical to the previous Tensor chip from a year ago. Alongside the two Cortex-X1 cores, Tensor G2 uses two Cortex-A78 at 2.35 GHz (the previous generation used two Cortex-A76): in single-thread tasks, the performance is destined to remain almost unchanged, while in multi-thread, the difference should be evident.
Alongside these first four cores, the Tensor G2 adds four more energy-efficient cores: the usual Cortex-A55 at 1.8 GHz. While Samsung and other well-known smartphone makers “rev their engines” to embrace Cortex-X3 cores, Google always prefers to stay a generation behind. Why? Because the Mountain View company prefers not to compete with competitors regarding raw specifications: Google wants to focus on the user experience by ensuring that its chip is optimized for processing related to artificial intelligence thanks to YOU ( Tensor Processing units ).
On the other hand, the name Tensor chose for its SoC comes from here: a TPU, in fact, is an accelerator developed for specific applications in the field of neural networks. TensorFlow is an open-source machine learning software library that provides “ad hoc” modules for handling tasks related to artificial intelligence algorithms. Since the processing is performed locally on the user’s single device, Google has chosen to call its smartphone chip Tensor.
The ARM Mali-G710 MP7 GPU that enriches the new Tensor G2 deserves a special mention: it is the latest generation of the graphics section. In the version that equips the Tensor G2, seven graphics cores were used while the previous Mali-G78 used 20 cores: the leap forward is clear in terms of architecture while the number of cores remains contained (think that a MediaTek Dimensity 9000+ uses a 10-core Mali-G710 GPU…). The “revamped” GPU should provide roughly the same performance as the previous generation.
In short, as Google managers abundantly underlined during the presentation, the absolute novelty of the Tensor G2 chip lies in the use of a primarily renewed TPU and in the optimizations that the company’s technicians were able to carry out by orchestrating the operation of the hardware on the one hand and the Android operating system on the other.
Processes related to machine learning are completed up to 60% faster than the previous generation of the Tensor chip and with an efficiency that can border on 20% more.
Google Tensor G2 significantly lowers energy consumption during regular daily smartphone use (video recording, multimedia streaming, voice recognition, instant translation of conversations,…).
From a security point of view, Tensor G2 interacts directly with the Titan M2 chip. Based on the RISC-V architecture, Titan M2 represents the smartphone version of the technology that Google has been using for years on the server side and is responsible for managing tasks related to security (such as the management of user information and personal data) with the possibility of connecting multiple CPU cores and threads.
Titan M2 also supports Android Strongbox, which generates and securely stores cryptographic keys to safeguard users’ passwords and PINs. In concert with the Tensor Security Core, it also protects user data keys during use in the SoC.
The Tensor Security Core is a custom security subsystem, a proper set of physical components that consists of a dedicated CPU, ROM, one-time-programmable (OTP ) memory, cryptographic engine, internal SRAM, and protected DRAM.