Google Tensor 2 expectations and what we want to see

With Google already lifting the lid on the upcoming Pixel 7 series so little, expectations and rumors about the phone’s next-gen Silicon are starting to spread. The upcoming chip, temporarily called Google Tensor 2, follows Google’s first expedition into the semi-custom silicon space and will power Google’s next-generation handsets.

Google’s next SoC could make a lot of changes, tweaks and complete improvements. After all, nothing is perfect on the first try. So let’s go through the current rumors and reasonable expectations about what will happen in our path after 2022.

Hold on: Everything you need to know about the first generation Google Tensor

Google Tensor 2 expectations and rumors

Early Google Tensor 2 rumors point to a few offerings for the next generation of chips. It will co-develop with Samsung, just like its predecessor, and will almost certainly be built on a new, smaller 4nm process. This is probably the same 4nm LPE process that Samsung uses for its own Exynos 2200 and Qualcomm’s Snapdragon 8 Gen 1 flagship silicon, or perhaps Samsung’s slightly newer and more efficient 4nm LPP process. What is less certain and open to a few possible variations is the makeup of the chip’s CPU, GPU and other components.

On the CPU side, if there is no combination and match between the Armv9 and Armv8 CPU cores, a choice is made between sticking to older components or moving to a higher-performance but potentially higher-priced CPU segment. Moving to off-the-shelf Armv9 components is a possible next step, as Google does not design custom CPU cores and we do not see X2 / A78 or any such combination. Upgrading the current large-arm Cortex-X1, the Medium Cortex-76, and the 55-cortex-Cortex-X2, Cortex-A710, and A510 to the current flagship Android smartphone of 2022 will provide a decent bump in performance. Especially since that would mean moving to a more modern mid-core than the dated A76 instead of 2018.

Read more: Arm Cortex-X2, A710, and A510 deep dive

Another unknown is whether Google will stand by its 2 + 2 + 4 cluster configuration or move to a more conventional 1 + 3 + 4 CPU setup. Either way, the decision will tell us a lot about whether Google thinks the system offers the right balance of performance and power consumption on its first-generation chipset.

Google Tensor 2 will pack an upgraded version of the company’s machine learning smarts. The current-General Tensor Processing Unit (TPU) combines the AI ​​processing capabilities of the chip with multimedia pipelines such as audio and image processing units. The close integration between these components allows Google to leverage its industry-leading AI algorithms for natural language processing, image recognition, and photo processing within the Pixel 6 series, which helps phones punch well above their price tag. See for example Magic Eraser, Pixel’s Signature HDR Image Processing and On-Device Speech Caption.

What is a SoC? All you need to know about the smartphone chipset

While we don’t know if Google will improve the TPU with the Tensor 2, we’re hoping for another tightly connected, unobtrusive piece of AI silicon to create a unique experience for the Pixel 7 series.

What we want to see

Stick to the TPU / Media Suite for a while, there’s room for improvement. We want to see Google Tensor 2 throughput reorganization of ISP to support enhanced video capabilities. For example, 4K 60fps video only works from the primary camera but it would be great to support the same feature for ultraviolet, telephoto and even selfie cameras. 8K recording can also be effective for more serious videographers. It’s not clear if this is a limitation of existing chipsets or camera modules, but we do know that Samsung’s Exynos chipsets have this capability and will be a very tempting feature for the Pixel 7 series. Similarly, the Pixel 6 has no simultaneous multi-lens capability, a feature we have seen in other smartphones. A wider, more powerful imaging pipeline would be the answer.

Even if you don’t want the hardware, opening a TPU to third party developers will be a feature of Google’s cap. The current TPU is very capable, but its potential is limited to Google’s in-house applications. Nice to meet you, of course, but there are many AI / ML applications that can also benefit from Google’s hardware acceleration.

See more: How on-device machine learning has changed the way we use our phones

Graphics is a field that has seen huge gains in 2022 and we want to see Google Tensor 2 equal to the playing field. Going from G78 to an arm Mali-G710 is a potential improvement, which could improve performance by 20% or more. However, this may not be enough to close the gap on the Snapdragon 8 Plus Gen 1 that will appear on handsets at the same time as the Pixel 7, but it has not yet been announced that the Snapdragon 8 Gen 2 will power the 2023 flagships.

Samsung’s partnership with AMD has produced a decent Xclipse 920 GPU on the Exynos 2200. If Samsung could license the results of its RDNA2 collaboration, it could lend itself to a suped-up Xclipse GPU chipset for Google Tensor 2. Serious gaming chops. While gaming isn’t Google’s priority, the competitive frame rate will definitely help Pixel 7 jack up all the business.

Finally, Primary code name Point to the Tensor 2 featuring a Samsung Exynos modem, which is to be expected. However, the codename still suggests the unannounced Exynos 5300 model. This may be an error at this stage, but it may also point to a semi-custom and / or integrated content explicitly produced for Google. The original Google Tensor used an external Samsung modem. Moving to an integrated component would be a boon for the power consumption and battery life of the device, especially when fast 5G pipelines are down data.

Read more: Snapdragon 8 Gen 1 vs. Google Tensor

Made for Google Pixel 7

The original Google Tensor was not a particularly ground-breaking SoC. Nevertheless, it was very innovative in the field of machine learning, which enabled Google to stuff its Pixel 6 and Pixel 6 Pro smartphones with impressive AI-based features. Instead of the out-and-out benchmark thrashing chip, Google and Samsung have designed Silicon uniquely to suit the look and needs of its mobile. We can almost certainly expect that this policy will continue to be the core of Google Tensor 2’s design.