Analysis | What is the level of Google Tensor?

preface:
Throughout the current mobile phone market, several giants, without exception, have their own SOC chips.
Even the rising stars such as Xiaomi, Vivo and OPPO are working hard on the road of core making. Head Google will naturally move towards the path of manufacturing SoC.
Author | Fang Wen
Picture source | Network
What is the level of Google Tensor?
Self-research or magic change, this is a problem
Tensor’s name comes from Tensor Flow, the famous AI learning framework of Google, which symbolizes the chip’s outstanding strength in AI. In addition, it is also to pay tribute to the tensor processing unit (TPU) launched earlier.
The new Pixel6 and Pixel6Pro phones launched by Google on October 20 are also equipped with Tensor chips.
But recently, some people found that the chip with the name of [independently developed by Google] is suspected to be changed from Samsung chip.
Google claims that the TPU, ISP and TitanM2 security chip of Tensor chip are its own technology.
But it has been revealed before that Tensor’s kernel code is S5E9845, which is very similar to Samsung’s Exynos 2100 S5E9840.
In other words, Google Tensor is probably the doll of Samsung Exynos, or the undisclosed Samsung Exynos 9855.
What is the level of Google Tensor?
Tensor is the first work of Google Silicon team.
In short, Tensor is close to Samsung Exynos in naming rules, and a few years ago, it was said that Samsung Semiconductor has begun to provide [semi customized] chip services.
Star will provide customized technology and functions according to customers’ needs, and can even intervene in the chip design stage.
In addition to background information, Tensor and Exynos use the same CPU and GPU architecture, and the power management, IO, storage controller and other major functional modules are of the same origin.
Tensor is a customer of Samsung (semi customized) chip service. Google provides the design theme, and Samsung is responsible for the construction and production.
Strictly speaking, Tensor should be a [customized] chip, not created from 0 to 1, but from 1 to 2, or 1 to 3.
What is the level of Google Tensor?
Tensor integrates self-developed TPU core
The Tensor processor developed by Google this time also integrates the TPU kernel developed by Google, but in terms of driver, Google calls it “edge TPU”.
Although Google did not introduce the performance indicators of Tensor’s built-in TPU, some data showed that the power of the TPU was 5W, which was different from the previous Google edge TUP.
Therefore, if the two are indeed related, the performance of Tensor TPU should be more powerful than the previous edgeTPU in view of the significant advantages of process nodes and the overall updated IP.
Therefore, if the two are really related, the performance data of Tensor TPU should be very important in view of the significant advantages of process nodes and the overall updated IP.
What is the level of Google Tensor?
Tensor chip is called the biggest innovation
Google CEO Sandal Picha called Tensor chip [the biggest innovation in Pixel field to date].
As for the big core, choose the 5nm A76 instead of the updated A78. The early generation Tensor still gives Google a cut in the choice of CPU and GPU. The 2+2+4 architecture, under the current scheduling of Android 12, still has a certain gap with Google’s expectations.
For Pixel, even without Tensor, Pixel6 and 6Pro are still Pixel’s products that turn a new chapter, return to mainstream hardware, new Android 12 and enough design for Google.
As Google Silicon team Mark said, under the unified processor architecture, there are too many traditional products in the market. Google Pixel series wants to share a piece of cake. Choosing different strategies is a good choice.
Tensor, whose absolute performance is not in the first tier, is actually successful for Google. It has not been hijacked by the market tide. Tensor’s design of AI and machine learning has marked the strongest mark of # Google # for Pixel series.
Tensor chip is a self-developed processor that took Google four years to officially launch.
For users of Google, Tensor chip can obtain better photo taking experience because of its creative core algorithm and image signal processing unit design.
Tensor’s interaction hub can better improve the existing speech recognition and input, real-time translation and many other personalized AI performance.
What is the level of Google Tensor?
Perhaps the significance is more than practical
As Google’s first self-developed chip, it may be more meaningful than practical.
According to the data released by Google, the Tensor processor is completely different from the Exynos 2100 with only one large core in architecture design.
In addition, the Tensor processor is different from the CPU and GPU, and the TPU, ISP and the matching TitanM2 security chip belong to Google’s own technology.
Tensor processor is born for computing photography. In addition to photography, Tensor processor can also improve speech recognition, voice commands, translation, subtitles and dictation, and even tap the maximum performance of the latest generation of Android system.
The design concept here also shows the changes in mobile phone use scenarios today: the CPU used for general computing is not suitable for more and more app use scenarios, and the huge power consumption will also affect the use experience.
Like Apple’s previous Mac, most of Google’s Chromebooks currently carry Intel’s low-cost x86 chips.
Samsung is no longer a simple chip manufacturer, but is fully involved in chip design, which can be compared with ASIC design services.
But this is a very special situation. After all, Samsung not only has chip OEM business like TSMC, but also has its own self-developed SoC.
Google Tensor and Samsung Exynos are highly homologous. In addition to the CPU, GPU, NPU and other advanced structures, many of the basic structures in the chip are homologous.
Although on paper, Samsung, MediaTek, HiSilicon, and even Qualcomm (only in terms of CPU) all use the Cortex CPU and Mali GPU public architectures of arm, their underlying architectures are very different.
In general, Google has indeed designed and defined Tensor. At the same time, there are many unique designs of Google, which are the differentiation of the whole chip.
What is the level of Google Tensor?
ending
If you study carefully, Apple’s early self-developed chips are also optimized and improved from the standard arm design.
As the first step of Google’s self-developed chip, we can’t expect the amazing performance of Tensor processor.
Part of the data reference: Mgke.com: “Can Google’s Tensor processor really” beat Qualcomm “by using two large cores at one time?” Lei Feng.com: “The actual performance of Google Tensor’s self-developed mobile chip specification analysis is not satisfactory.”