Tensor against Snapdragon 888 and Exynos 2100; Can Google’s first mobile chip compete?

تنسور در برابر اسنپدراگون ۸۸۸ و اگزینوس ۲۱۰۰؛ اولین تراشه موبایل گوگل توان رقابت دارد؟

Google unveiled the long-awaited Pixel 6 series phones last week; Products that use the first mobile chip of this internet search giant. This chip, called the tensor, has to compete with the big names in the industry, so what does the tensor do against the Snapdragon 888 and Exynos 2100? In this article, we want to compare Google Tensor with Qualcomm and Samsung flagship chips.

Google could get the chips it needed from Qualcomm this year, as it did in previous years, or even go to Samsung and buy a chip from the world’s largest maker of Android phones. But the company made another decision and entered the market personally, although such entry was not without help.

Google partnered with Samsung to develop the tensor chip, and Silicon incorporated its own machine learning. This chip is slightly different from the flagship chips of Android 2021. Although we still have to wait for a full review of the benchmarks and power consumption by the tensor, we now have enough information to be able to compare it at least on paper with Samsung and Qualcomm chips.

The Snapdragon 888 is used in many of the flagship Android 2021 handsets, and Samsung has used the Exynos 2100 in some Galaxy S21 series models. Therefore, in this article, we want to compare the tensor with these chips to see if it can theoretically compete with the market leaders.

Compare Google Tensor with Snapdragon 888 and Exynos 2100

Although we are not far from the introduction of the next generation of flagship chips from Samsung and Qualcomm, Tensor has come to compete with their current generation, the Exynos 2100 and Snapdragon 888. The following table shows the specifications of these chips:

Google Tensor Snapdragon 888 Exynos 2100
CPU 2 Cortex-X1‌ cores clocked at 2.8 GHz + 2 Cortex-A76‌ cores clocked at 2.25 GHz + 4 Cortex-A55 cores clocked at 1.80 GHz 1 Cortex-X1 core at 2.84 GHz and 3 GB for the Plus model + 3 Cortex-A78 cores at 2.4 GHz + 4 Cortex-A55 cores at 1.8 GHz 1 2.90 GHz Cortex-X1 core + 3 2.8 GHz Cortex-A78 cores + 4 2.2 GHz Cortex-A55 cores
GPU Mali-G78 MP20 Adreno 660 Mali-G78 MP14
Machine learning Tensor processing unit Hexagon 780 DSP Triple Neural Processing Unit + DSP
دیکد مدیا H.264, H.265, VP9, ​​AV1 H.264, H.265, VP9 H.264, H.265, VP9, ​​AV1
Modem 4G LTE and 5G below 6 GHz and millimeter wave 4G LTE and 5G below 6 GHz and millimeter wave – download speed of 7.5 Gbps and upload speed of 3 Gbps (Snapdragon X60 integrated modem) 4G LTE and 5G below 6 GHz and millimeter wave – download speed of 7.35 Gbps and upload speed of 3.6 Gbps (Exynos 5123 integrated modem)
Lithography 5 nanometers 5 nanometers 5 nanometers

As can be expected from the deep connection between Samsung and Google, the tensor is highly dependent on the technology in the latest Exynos chip. For example, we expect to see the use of the Exynos 2100 modem in the Google chip. In addition, both of them use the Mali-G78 graphics processor, although in the Google chip we see the use of 20 graphics cores and in the Samsung chip we see 14 graphics cores. They even share hardware support for AV1 Media Decode.

We can not comment on the graphics performance of Google Chip, however, it should not be much different from the Snapdragon 888, and of course its performance will be higher than the Exynos 2100. In general, we are dealing with an acceptable flagship chip that provides high power to buyers of Pixel 6 series phones. We even expect the tensor processing unit (TPU) of this chip to severely challenge competitors in terms of machine learning and artificial intelligence capabilities.

The Google Tensor chip is apparently a worthy competitor to the Snapdragon 888 and Exynos 2100 in terms of processing, graphics, modems and other technologies.

Google has gone to different and almost strange cores to design the CPU and we are faced with 2 + 2 + 4 configuration. The Internet search giant used two Cortex-X1 cores, which indicates the performance of a suitable single-core, but for the next cluster, it used the old Cortex-A76 cores, which can negatively affect its performance in the multi-core segment. We have to wait and see how this combination performs in terms of power as well as thermal management.

On paper, it seems that Google’s tensor chip in the Pixel 6 series handsets can challenge the Exynos 2100 and Snapdragon 888 well, and thus the Android 2021 flagships.

Google tensor chip CPU design

An important question about the tensor chip architecture is the use of 3-year-old Cortex-A76 cores. Why did Google go for this core instead of Cortex-A78? The answer to this question relates to chip size, power and thermal management.

ARM compares their performance to their predecessors by introducing new processing and graphics cores. As you can see in the slide below, the Cortex-A76 is smaller and uses less energy than previous generations, yet has the same frequency and manufacturing process. The following slide is based on 7nm lithography, but Samsung has been working on the 5nm Cortex-A76 for some time.

1635189661 934 Tensor against Snapdragon 888 and Exynos 2100 Can Googles first Tensor against Snapdragon 888 and Exynos 2100; Can Google's first mobile chip compete? 2

If you are looking for numbers, the Cortex-A77 is up to 17% larger than the Cortex-A76, while the A78 is only 5% smaller than the A77. In terms of energy consumption, there is a big difference between these cores, and the A78 consumes only 4% less energy than the A77. Therefore, the A76 is a smaller and less energy-efficient core than them.

So far everything has been positive, but let’s look at the downside of such an approach. The maximum performance of the A77 core is up to 20% higher than the A76, and the difference between the A77 and the A78 is up to 7%. Therefore, the multi-core tensor should perform worse than the Snapdragon 888 and Exynos 2100, and early benchmarks suggest this, but it also depends on performance. Given the use of two powerful Cortex-X1 cores, Google is likely to be confident in the optimal performance and power consumption of its chips.

The use of older Cortex-A76 cores is highly dependent on equipping the tensor with two powerful Cortex-X1 cores. Of course, there is a challenge in this, in the design of mobile chips, manufacturers must pay attention to its dimensions, power and heat output, so the use of two Cortex-X1 cores challenges such a thing.

The use of smaller and less energy-efficient cores gives the manufacturer more space and better heat management capabilities. So the use of two powerful Cortex-X1 cores has forced Google to move to older cores. But there’s an important question: Why, while Qualcomm and Samsung are satisfied with the performance of a Cortex-X1 core in their chips, has Google gone for two cores?

To answer that question, Google’s vice president and general manager Phil Phil Carmack said such an arrangement was created for better performance for less powerful activities. According to Carmack, a strong core is unique to a single-core benchmark, but two powerful cores are the best way to achieve high performance.

Google Tensor

In addition to the high-performance single-core performance of the Cortex-X1, it is up to 23% faster than the Cortex-A78 and has a high level of machine learning performance. This could be the reason for the use of two Cortex-X1 cores in the tensor chip, as Google pays special attention to machine learning and artificial intelligence.

The Cortex-X1 kernel is twice as good at machine learning as the Cortex-A78, due in part to its higher cache and twice the bandwidth of the SIMD floating point instruction. In general, instead of multi-core performance, Google has moved to two large Cortex-X1 cores to enhance TPU machine learning performance. We do not know at this time how much cache Google has provided for these cores, but it could have a big impact on its performance.

Two Cortex-X1 cores improve the machine learning performance of the tensor chip

Despite the use of low-power and small Cortex-A76 cores, it is still possible that Google’s tensor has a problem with power and heat. Experiments have shown that the Cortex-X1 core consumes a lot of energy and can not maintain its maximum frequency well in today’s flagships. Some smartphones do not perform activities as much as possible, even in order not to increase energy consumption.

Now Google has decided to use two Cortex-X1 cores, which doubles the power consumption and heat generation problem, so the 6-pixel series tensor chip cannot easily be considered the winner. It is possible that this chip will have problems in such areas, although Google apparently does not believe so.

The company considers the use of an older, less energy-efficient core and its less powerful activities to be a good reason to use the two Cortex-X1 cores. According to Google, such a configuration is a good point between performance and energy efficiency. However, the chip must be tested in the real world, as well as see how it performs in heavy activity. However, according to initial benchmarks, the tensor chip performs worse in the multi-core segment than Qualcomm’s Snapdragon 888 and Samsung’s Exynos 2100.

Google TPU

Google Tensor

One of the unknowns in the tensor chip is its tensor processing unit or TPU. We know that the task of such a unit is to perform machine learning activities from voice recognition to image processing and even video decoding. So we seem to be dealing with a general application interface and media pieces for media tensor functions.

Qualcomm and Samsung have also used dedicated parts on their chips for machine learning functions. Meanwhile, Qualcomm’s performance is interesting because it has placed processing units in different parts. Qualcomm’s AI engine is housed in the CPU, GPU, Sensing Hub, Hexagon DSP and Spectra ISP. Although such an approach is suitable for energy efficiency, it cannot use all the components at the same time. Therefore, Qualcomm’s 26 TOPS AI performance is not very accessible. In fact, most of the time, the two parts work together at the same time, like ISP and DSP for computer vision tasks.

Google’s TPU unit will include various sub-blocks, especially if it encodes and decodes videos, however, it seems that a large part of the 6-pixel machine learning capabilities depend on it. If Google can use the maximum power of TPU at the same time, it can surpass competitors and make a big leap in AI and machine learning performance.

Applications of this unit, according to Google, include offline voice dictation, offline voice translation, blurring of faces in images and 4K video recording at 60 frames per second with hardware-specific mode in the 6-pixel series chip called “HDR Net”. کرد.

Comparing Google Tensor with Snapdragon 888: Preliminary Summary

Tensor against Snapdragon 888 and Exynos 2100 Can Googles first Tensor against Snapdragon 888 and Exynos 2100; Can Google's first mobile chip compete? 6

With the withdrawal of Huawei crane chips from the market due to US sanctions, a new spirit in the chip market is emerging. Although according to the preliminary results of the benchmarks, this chip can not compete with the market leaders even in the Android world, but it can have a very bright future.

With the tensor chip, Google can not make a leap in the performance of the current generation of chips, but apparently with a new approach is looking to solve some problems with mobile chips. With two powerful processing cores as well as TPU-based machine learning functions, the company is slightly different from its competitors. In addition to these, we should mention the 5 years of security support for the Pixel 6 phones, which is possible thanks to the tensor.

What do you think about Google Tensor? Do you consider Google’s first attempt at the smartphone market to be successful, or is it a long way from the Snapdragon 888 and Exynos 2100? Share your opinion with us and other users.

More Similar Posts

Leave a Reply

Your email address will not be published. Required fields are marked *

Fill out this field
Fill out this field
Please enter a valid email address.
You need to agree with the terms to proceed

Most Viewed Posts
Techno barg We would like to show you notifications for the latest news and updates.
Allow Notifications