Google Pixel 6 is evolving into the company’s most ambitious smartphone in years, thanks in large part to a new, customized Tensor processor that aims to put Google at the forefront of the smartphone market with Google’s years of machine learning experience. And Google needs it: despite Pixel’s popularity in technology circles, its phones just aren’t popular sellers in the U.S., barely moving the needle compared to Juggernauts like Samsung and Apple or even smaller players like Oppo or Xiaomi.
Tensor is Google’s big bet, focusing on an artificial intelligence-enhancing TPU that promises to enhance photos and videos, search, subtitles, text-to-speech, and more. It’s a high subscription to any chip, much less focused primarily on machine learning as its special feature – but while Tensor SoC may not launch Google Heights like the iPhone however, it can be a crucial first step towards turning Android into a top competitor as well.
The rest of the SoCs are a mystery at the moment, but it looks like Google is using third-party models for things like CPUs, GPUs, and modems — which means Pixel 6 probably feels the same as any other Android smartphone from Qualcomm or Samsung’s CPU allows for most tasks instead of some kind of revolutionary upgrade with Apple’s A-Series iPhone.
Google has not provided much information about what the actual architecture of Tensor will look like outside the CPU, GPU, modem, or other key components of the SoC outside the TPU. But based on rumors and the fact that Google won’t take the opportunity to ask about any major customizations or advances it makes, it’s likely that most of Tensor’s hardware stack is outsourced models. Qualcomm and Samsung are already doing something similar – the Snapdragon 888 uses partially customized versions Arm’s Cortex-X1, A78 and A55 models, while the Samsung Exynos 2100 uses Arm models for both CPU and GPU.
XDA notes more specifically, the Tensor is likely some sort of combination of Arm’s Cortex-A78, Cortex-A76, and Cortex-A55 processor cores, as well as Arm’s standard Mali GPU. This means that the difference between a Tensor and a Snapdragon 888 or Exynos 2100, for example, may not be as great in terms of overall CPU or GPU performance, for example. This is a good category, especially if Google is really trying to make a flagship device.
But like Google Rick Osterloh said Limit, “Ordinary goods that people look at are very competitive and artificial intelligence matters stand out completely.” However, what makes the Tensor special and unique is not how fast it can run games or how efficient its processor and battery are.
This means that Tensor is probably not the magic solution Android fans have been hoping for either: a custom chip made by Google specifically designed for Android and Pixel hardware to deliver the performance and power that Apple has been able to provide on the iPhone for years with a similar strategy .
But the good news is that Tensor is just a first generation product; It’s easy to point to Apple’s custom chips and require Google to do the same, but it’s important to remember that Apple’s earlier A-Series iPhone chips started the same way. The A4 and A5 chips had standard Arm models of CPU cores (with some optimizations and improvements on top) before Apple switched to fully customized models in later generations.
Google is rumored to be working with Samsung, both of which make chips – which makes sense given that it is one of only two companies to work with TSMC, which currently makes 5 nm chips (a service it offers to a number of companies). , including Qualcomm).
Samsung is also rumored to work more closely with Google on chip design, including leveraging its Exynos hardware and software outside of manufacturing alone. We’ll have to wait until Google reveals more about Tensor to know the scope of the partnership, but that could mean Google gets more help improving the more essential custom chips than if it had to reinvent the wheel from scratch.
The idea of a purpose-built machine learning circuit on a Pixel phone is also not a new idea for Google. The company has previously added its own artificial intelligence circuits to its phones, such as Pixel Neural Core and Pixel Visual Core In the Pixel 2, PIxel 3 and Pixel 4 models, which makes the Tensor less new and more advanced work.
The Tensor TPU appears to be the next step in this, and presumably offers more powerful artificial intelligence functions than either of the previous two chips. It’s also more tightly integrated with the phone, and Google points out that it can do, for example, route image data directly through the TPU.
These benefits may be worth developing our own custom SoC – once again we have to wait and see how Google actually implements these artificial intelligence features and how significant an improvement they are compared to the standard Android flagship. (A feeling that usually relates to how strongly Google emphasizes TPU as the main differentiator of Pixel 6, which feels awkward to actually sell to customers.)
However, the current main difference between the Tensor and the Snapdragon 888 seems to be that it’s just great neural network tricks. But just like with Apple’s early internal chips, Tensor may just be the first step on a longer journey toward a custom Google chip. And it’s a much more exciting concept than subtly better machine learning demonstrations.