Tensor G3, which shouldn’t be released until the fall with the new Pixel 8 series smartphones, is produced in partnership with Samsung’s semiconductor division, as in previous versions.

Performance, security, new kernels

According to Android Authority, the source of the information is someone from the company itself.

The Tensor G2 was a fairly uninteresting chipset in terms of CPU performance. At the time of release, all cores were already two generations behind competitors. The only real change compared to the first generation chip was the replacement of the archaic Cortex-A76 cores with the more modern Cortex-A78. The chip retained the unusual 4+2+2 core layout, while most other manufacturers used the single large core 4+3+1 layout.

In the Tensor G3, Google changes the architecture of the entire processor unit and finally uses more modern ARMv9 cores. The location of the cores is now new – the unusual 4+2+2 order is gone. Instead, Google put something weirder. Tensor G3 will have nine processor cores:

• Four small Cortex-A510s with an operating frequency of 2.15 gigahertz.
• Four medium Cortex-A715s with an operating frequency of 2.45 gigahertz.
• One large Cortex-X3 core with a frequency of 3 gigahertz.

With working frequency will increase compared to previous generations. This should lead to a significant increase in performance and will allow the Tensor G3 to match the quality of other flagship SoCs of 2022. However, it will still lag behind chips using the recently announced ARMv9.2 cores.

Switching to ARMv9 also new security technologies. The Pixel 8 will have Arm’s Memory Tagging Extensions (MTE) that can prevent some memory-based attacks.

The main change in ARMv9 is that only 64-bit architecture. Tensor G2 devices, such as the Pixel 7 series, have already dropped support for legacy 32-bit applications, though they retain built-in 32-bit libraries.

It is also expected a significant increase in graphics performance Thanks to the switch to the Arm Mali-G715. This will be useful not only for demanding games, but also for the artificial intelligence work that Google devoted almost the entirety of its May presentation to. The company will apply AI in literally every possible space, app, and device. There is no information about possible ray tracing.

Tensor G3 also updates video features. The camera can now shoot video in 8K format, but the source says Google Camera is unlikely to get the feature. The current test version doesn’t seem to have this feature, and also the Pixel already has heating issues when recording 4K, and increasing this figure will only make things worse, not to mention how much memory this type of video will take up.

Artificial intelligence

Tensor’s main focus is clearly artificial intelligence.

• The first-generation Google Tensor comes with a built-in TPU codenamed “Abrolhos” that operates at 1 gigahertz. It provided excellent performance, especially in natural language processing (NLP) tasks.

• The Tensor G2 has upgraded the TPU to the codename “Janeiro”, which still runs at 1 GHz. Google claims it’s 60% faster than the original chip in camera and speech recognition tasks.
• Tensor G3 is expected to include a new version of TPU codenamed “Rio” running at 1.1 GHz. Although there is no concrete data on its performance at the moment, it is certain that the Rio will be a significant upgrade.

What else

• A new dedicated DSP processor is also expected for tasks such as image processing, artificial intelligence rendering and the Magic Editor function in Google Photos, which will be necessary for the camera to work even better.
• Faster UFS 4.0 memory that doubles the speed and increases efficiency by 50%.

Unlike the Tensor G2, which is a minor upgrade, the Tensor G3 looks like a big step forward.