Google’s Pixel 7 and Pixel 7 Pro feature a new system on a chip (SoC) inside, designed in-house. It’s the second generation Tensor SoC – called Tensor G2 – and it’s named after the Tensor Processing Units (TPU) that Google uses in its data centers. Here’s what you need to know about Tensor.
What is Google Tensor?
Tensor is Google’s in-house SoC. It’s not a single processor. Before the Pixel 6 and Pixel 6 Pro, Google relied on Qualcomm processors, as do most other Android devices in the US. Globally, however, there is more variation, with Samsung and MediaTek also producing chips for popular Android phones.
Keep in mind Apple has its own in-house silicon, the A-Series line of chips. That doesn’t mean Tensor and Apple’s A-Series are equivalent – Tensor is a system on a chip using a mix of components that Google has either designed or licensed. That’s perhaps why Google is spending so much time hammering home Tensor’s artificial intelligence and machine learning advantages, rather than discussing specifics about CPU, GPU, and RAM.
Those components impact how fast a phone seems, how long the battery lasts, cellular connectivity, etc. There are also co-processors that handle dedicated tasks such as image processing or security. Google does have experience in those areas: the Pixel Visual Core and Titan M chip, respectively, both have appeared in previous phones. And don’t forget Google has also been making Tensor Processing Units for its servers for years.
Tensor even refers to Google’s open-source TensorFlow machine learning programming ecosystem. But Google Tensor marks the first time Google has introduced a mobile TPU specifically for a phone. Ideally, it allows a phone to process more information on-device, rather than send data to the cloud.
So excited to share our new custom Google Tensor chip, which has been 4 yrs in the making (for scale)! Tensor builds off of our 2 decades of computing experience and it’s our biggest innovation in Pixel to date. Will be on Pixel 6 + Pixel 6 Pro in fall. https://t.co/N95X6gFxLf pic.twitter.com/wHiEJRHJwy
— Sundar Pichai (@sundarpichai) August 2, 2021
What can Google Tensor do?
Google provided press demos of what the new TPU inside Tensor could do in the Pixel 6 months before the first chipset actually unveiled the phone. The Tensor G2 is very similar, but features an improved machine learning core. Google showed it improving and sharpnening old, blurry photographs in one demo, and those photos weren’t even taken on the Pixel 7.
The initial reveal of the Pixel 6 and its Tensor chip was largely focused on the AI-enhanced TPU, and the company doubled down on that angle when it launched Pixel 7 in October 2022. It showed features like the Real Tone photography and a Guided Frame feature which helps those who are blind, or with low visibility, to take a well-frame selfie through accurate spoken prompts.
Google said Tensor can handle quick Google Assistant queries and longer audio tasks like live captions. In another demo, Google teased an on-device auto-caption feature that quickly and accurately translated language in a playing video from French to English in real time. And has since shown its Recorder app being able to work in multiple languages, and being able to detect different speakers automatically. All processed on device, securely.
Google – in the past – has also showed how Tensor enabled other Pixel features, such as Motion Mode and more accurate face detection. There’s also a context hub that it powers. This is used for ultra-low-power AI tasks, like the now-playing alerts that show on the always-on display when the phone is locked.
What are Google Tensor’s specs and benchmarks?
From a purely technical standpoint, the Tensor G2 isn’t all that different from the first generation Tensor.
Tensor G2 packs a next-gen custom TPU (Tensor Processing Unit) for AI with a CPU cluster of two Arm Cortex-X1 cores, two mid-range (older Cortex-A78 cores), and then four low-power efficiency cores (Arm’s Cortex-55). Graphics are powered by an Arm GPU. Plus, there’s an image processing core.
It’s not as big a leap in machine learning performance as the Pixel 6 was from the Pixel 5, but it is an improvement nonetheless.
In addition to the TPU, the Pixel 7 sports Google’s Titan M security chip, to help keep your personal data secure and sensitive, personal requests processed on device. .
Is Google Tensor ARM-based silicon?
Yes. According to Google hardware chief Rick Osterloh (via Wired), Tensor is based on the same ARM architecture as Qualcomm and other mobile processors.
Which devices offer Google Tensor and Tensor G2?
Google released the first Tensor-powered Pixel phones – Pixel 6 and 6 Pro – in October 2021. Those were followed by the more affordable Pixel 6a in 2022. All three powered by the first generation Tensor chipset.
The first devices powered by the second gen Tensor – the Tensor G2 – will be the Pixel 7 and Pixel 7 Pro, which were launched in October 2022. They’ll be followed in 2023 by the first Tensor-powered Pixel Tablet.
Who developed Google Tensor?
Google said Tensor was co-designed alongside Google Research. Samsung is fabricating the chips on its 5-nanometre LPE process.
Why did Google go its own way?
It’s been six years since Google started making Pixel phones, yet it’s still often asked whether it’s serious about hardware. Google said it began work on Tensor silicon five years ago (around the time it announced it wanted to buy HTC’s phone hardware division). We suspect, with all of Google’s main competitors gaining experience making their own in-house silicon, Google wanted to prove the Pixel line isn’t just a hobby project for it.
But Google has also talked a lot about how computing limitations inspired it to make Tensor. “We set about building a technology platform built for mobile that enabled us to bring our most innovative AI and machine learning (ML) to our Pixel users,” Google said in a blog post announcing Pixel 6. “We set out to make our own System on a Chip. . . Tensor was built for how people use their phones today and how people will use them in the future”.
The company explained that, increasingly, more smartphone features are powered by AI and ML. Tensor will not only unlock computing resources for those tasks, but it can also help provide new, specific experiences for Pixel users, whether it’s a revamped camera system or speech recognition.
The highlight of Tensor is that it can process Google’s most powerful AI and ML models directly on #Pixel6.
You’ll see a transformed experience for the camera, speech recognition and many other Pixel 6 features.
(6/13)
— Made By Google (@madebygoogle) August 2, 2021
Want to know more?
Check out Google’s blog post about Tensor.