Google Tensor G2 chip explained!

Introducing the Pixel 7 and Pixel 7 Pro, Google has not only unveiled its latest smartphones but also showcased the successor to its bespoke Google Tensor chipset. Named the Tensor G2, this chipset is the result of Google’s ongoing collaboration with Samsung Semiconductor for semi-custom silicon development. Powering various devices released in the first half of 2023, including the Pixel Tablet and Pixel Fold, the Tensor G2 follows a design philosophy that prioritizes specific user experiences over achieving peak performance.

Similar to its predecessor, the original Google Tensor, the Tensor G2 does not aim to top benchmark charts. Instead, it focuses on enhancing imaging, machine learning, and security features to provide users with optimized functionalities. While the Pixel 6 series, powered by the original Tensor, received positive feedback for its tailored experiences, there were some concerns regarding temperature and networking performance.

The Tensor G2 continues with the semi-custom approach, incorporating older off-the-shelf components at its core. Despite not being as snappy or efficient as the latest market components, the chipset relies on Google’s next-gen custom AI and imaging silicon to compensate. The Pixel 7 series places less emphasis on raw power, with a greater focus on specially crafted hardware to elevate Google’s bespoke speech, contextual support, imaging and video, and AI experiences.

As we delve into the Tensor G2’s inner workings, it becomes essential to scrutinize whether Google has succeeded in refining its approach in this second attempt and what users can expect from this new chipset in terms of performance and functionality.

Google Tensor vs Tensor G2 specs

First-gen Tensor vs. Tensor G2

The above table highlights that the differences between the original Tensor and Tensor G2 are minimal, with only a few core changes, and their significance may be limited.

To begin, the older Cortex-A76 middle cores, dated back to 2018, have been swapped with the more recent Cortex-A78 from 2020. These offer increased performance with slightly more space and power usage. The CPU setup remains mostly the same, featuring powerful but aging Cortex-X1s for heavy tasks and energy-efficient Cortex-A55s for background processes. While not a complete overhaul, this results in a modest performance boost, especially for gaming and sustained workloads.

The GPU configuration sees a revision, utilizing Arm’s 2021 Mali-G710 microarchitecture for a 20% boost in performance and power efficiency, along with up to a 35% improvement in machine learning capabilities. Google opts for a smaller 7-core setup, departing from the 20-core arrangement in 2021, potentially influenced by the varying performance levels of these new cores. Real-world performance results will be discussed in the following section.

In terms of innovation, Google introduces its latest Tensor Processing Unit (TPU) in the Tensor G2. Integrated into the image pipeline, the TPU efficiently handles various machine learning tasks, enhancing tasks like real-time translations, image, and video processing. Although specifics about the new TPU are not detailed, Google mentions a significant boost in speed—up to 60%—for camera and speech tasks. Additionally, the ISP now supports 10-bit HDR video recording, Google HDRnet tone mapping, and the capability for up to 108MP zero shutter lag images, despite the Pixel 7 series having a 48MP camera.

For more such updates, tips and learning resources, stay tuned to Insitebuild Blog.