Google Tensor G2 vs G1

The recently released Google Pixel 7 and 7 Pro have already excelled in several categories compared to

predecessors.Among the improvements visible to the naked eye, the new owners note two things. This is faster shooting using scenarios (there is no need to talk about quality, this is taste) and, oddly enough, more stable and sensitive signal reception from radio towers. To understand the essence of what is happening, there is nothing better than comparing two generations of chipsets, Google Tensor G1 and G2. The main changes are visible in the table:

It needs to be taken apart piece by piece.

Nuclei

Starting with the first Tensor chip, Google constructorsin collaboration with Samsung continue to amaze with their unconventional approach. While all other manufacturers, including leaders MediaTek and Qualcomm, are building their pyramid of cores with just one productive core of the ARM Cortex X series, Pixel smartphones have come up with a scheme with two such productive cores at once. I'm sure other manufacturers will follow suit with the development of true threading in Android software. But so far, only Google has had the courage for an unorthodox design, and other chipmakers are still gaining the highest scores in tests, overclocking the old design when using massive cooling systems. The scheme of Tensor cores looks like this:

At first glance, it seems that Google is trying“bring up” old technologies to the modern level, but there may be several reasons for this behavior. One of them has already been announced - the development and testing of real two+ threads of software, including system software. Google writes not only its services, but also Android itself, it is worth remembering this. Other reasons can be very different, from cost savings to such a banal thing as the consent of only one chipmaker (Samsung) to share development experience and part of the technology. Be that as it may, someone has to be first, and Google has taken this burden on its shoulders. At the same time, it is obvious that old user tests, such as Antutu or Geekbench, are not able to show the benefit of using two productive cores, they all float in the wake of mass products, where a “pyramid” is used with one productive core at the top. The same goes for the TPU block that Tensor G2 has worked so hard on. The Google Tensor G2 chipset itself is too similar to the Samsung Exynos 2100 5G to be considered a coincidence. The difference lies only in the "extra" productive core.

Apparent lagging behind industry leadersalso applies to production technology. Modern chipsets are made on 4 nm nodes, while both Google Tensor chips use Samsung 5 nm nodes, apparently as a result of cooperation between the companies. The problems of Samsung factories during the transition to 4 nm are also evidenced by the fact that in the spring of this year, Qualcomm moved the production of Snapdragon 8+ Gen1 chipsets to TSMC factories, and in some specialized publications this fact was even called "rescue operation" and "evacuation". Well, there are subtleties and even "generations" in the production of 5 nm chipsets from Samsung, but we do not have an electron microscope to check it.

Miscellaneous

Affiliate material

Reality and prospects of the IT professions market

What professions are the most popular and highly paid?

Saturday coffee #223

Pour a cup of invigorating Saturday coffee andcheck out the news of the week. Apple released new devices, Anker introduced sports headphones, Lenovo showed interesting concepts, and a new brand of cars will appear in Russia ...

Volkswagen Taos test. Nice compact crossover

Volkswagen Taos is the most compact crossover of the German automaker, which came to Russia in the summer of 2021, expanding the company's model range.

Sonos Roam portable speaker review

Smart functionality in a compact package…

Video card

Video card is an established term, and for itone word is enough to write, it is convenient, unlike the "graphics subsystem". The ARM Mali-G78 MP20 video card was installed in Google Tensor G1, respectively, the GPU consists of 20 cores out of 24 possible. This is a powerful graphics card, especially at the time of its introduction (2020). But this also has its own specifics, ARM recommends such chips, which are close to the limit, for use in devices with a good cooling system, which do not include smartphones. Numerous tests of Pixel 6 series smartphones have shown that the monster Mali G78 MP20 shows maximum power only for the first 5-10 seconds, and then its power drops following excessive overheating and power consumption. This story from the category of “what will happen if you put an engine from KamAZ into the Tavria” looks great on paper, but in practice this power is almost never used in modern heavy games due to the limitations of the laws of physics.

In Google Tensor G2, constructors have gone overin the traditional way and integrated the ARM Mali-G710 MP07 video card. There is no exact information, but most likely we are talking about 7 GPU cores out of 16 possible. This is already “not the limit”, quite an average value, which gives the video card the opportunity to work for a long time at close to maximum frequencies without overheating and increased power consumption. This is a logical decision, if you remember that Google Pixel smartphones have never been positioned as flagships. The manufacturer (Google) promises a significantly increased performance when executing Vulcan instructions, which is sure to please gamers.

In practice, the backlog of smartphones Google Pixel 7and 7 Pro from flagships is more than obvious. In the popular Genshin Impact game, the Tensor G2 behaves the same as the "spirit guide" Exynos 2100, showing maximum performance only for the first 30-40 seconds, and then cuts down on its ambitions and FPS.

Performance

When it's not about using functions withusing AI, both series of Google Pixel 6 and 7 smartphones based on Tensor G1 and G2 chipsets show comparable performance, the difference in test results does not exceed 3-8%. In the benchmark charts, both devices "hang" at the level of the Samsung Galaxy S22 and smartphones with Qualcomm Snapdragon 870-888, occupying a place between the flagships and all the rest. These are excellent results for 2021, but for the end of 2022 they can be called “above average”, no more.

The table shows the weak point of all Google Pixels - slow memory compared to flagships. This gives the flagships extra points in Antutu, but it has minimal impact on everyday use (UX).

The hardware update of the line is associated with optimization, but not with any breakthroughs.

Connection

This is a very tricky question, becauseto use 5G networks, 5G towers themselves are needed, the distribution of which has not yet reached the “everywhere” level. The same applies to modern public Wi-Fi networks, they are in the capitals, but two meters away they no longer exist. The only thing that the testers paid attention to (from numerous reviews in open sources) is that the Pixel 7 is better than the Pixel 6 at picking up the base station signal. The reason, as always, lies in the hardware. The Google Tensor G1 chipset integrates Samsung's first-generation 5G modem, the Exynos 5123. This is indeed an ancient modem that came to us from the era of the Samsung Galaxy S11, with a long gray beard and memories of the first 5G assault. In Tensor G2, a more recent version was installed - Exynos 5300. Again, I repeat, it will be difficult to assess the advantage in our country, we can only hope for a guaranteed bonus - reduced power consumption in all operating modes, which certainly happens with the release of each new generation of modems for smartphones.

TPU

First encountered with the abbreviation TPU (GoogleTensor Processing Unit) is worth going to Wikipedia at this link, but in simple words, let's say that this is an additional chipset unit responsible for processing graphics as quickly as possible, but with reduced accuracy. It can also be called an additional specialized video card, if you like.

The core value of a dedicated TPU block is in the chipsTensor (and others) is directly related to photography and video. If the Pixel 6 series smartphones boasted a quality bokeh effect, then the Tensor G2 in the Pixel 7 takes on even more complex tasks. These include bringing bokeh to video (called Cinematic Blur) and anti-blurring for photos called Photo Unblur.

Photo Unblur example

The value of the second function is especially great, itturns blurry faces of relatives and friends in hastily taken photographs into something more or less acceptable. These functions are needed by normal people no more than once a year, but this argument did not stop Google, she invested a lot of money in the development of a new TPU block, which she simply called “next generation”. As if the old generation in the Google Pixel 6 was bad.

Conclusion

The only question that may concernelectronics lover, it sounds like “is it worth switching from Google Pixel 6 to Pixel 7?”. Let's break it down into pieces. The Pixel line has already emerged from the slender ranks of typical Android smartphones, but is still far from separating into a separate category, like the iPhone in relation to everyone else. At any time and without any problems associated with addiction, you can leave the world of Pixel in the warm embrace of any other manufacturer of robophones. What is the difference between Pixel 7 and Pixel 6, apart from design and screen size? Faster work of branded functions from Google, a couple of new "pop" features, but nothing else.

The terminally orthodox smartphone lover is notwill see such a significant difference between the last two generations of Pixel to pay for it. As he will not see much difference between Tensor G1 and Tensor G2, which is more marketing than actual.

What do you think?