Das selbstentwickelte SoC von Google wird zum Witz in Bezug auf die Leistung. Spielt Google jedoch hinter den Kulissen eine größere Strategie?

If we were to rank the worst self-developed SoCs in the smartphone industry currently, Google would definitely be near the top of the list. There's no other reason. Objectively speaking, based on benchmark scores, Google's self-developed "Tensor" SoCs in recent years have been shockingly poor in both architectural design and performance metrics.

Initially, people thought it was "Samsung's fault." As is well-known, the four generations of Tensor SoCs from G1 to G4 were essentially designed by Samsung, with only some Google IP added. From their design concepts (such as dual super-large cores and ten cores), we can indeed see some typical features of Samsung's Exynos series of flagship SoCs during the same period.

However, when the "Google self-developed, TSMC-manufactured" Tensor G5 appeared in the Pixel 10 series in the fall of 2025, everyone realized that they might have wrongly blamed Samsung.

The "obvious" fact is that the overall performance of Tensor G5 has shown little significant improvement compared to the Samsung-designed Tensor G4. In fact, some aspects have even regressed significantly.

The rumored AnTuTu benchmark scores of Tensor G5. Left is the score when it was just released, and right is the score after a recent system update.

Meanwhile, Samsung's latest Exynos 2500 has performed far better than Tensor G5 in actual tests. This even implies that Samsung is capable of manufacturing high-end SoCs. The poor performance of previous generations of Tensor SoCs was either due to Samsung "holding back" or Google's mismanagement.

So, what "mistakes" did Google make that led to the poor performance of Tensor G5? According to AnTuTu benchmark scores, in the latest system version, Tensor G5's overall score is just over 1.3 million, similar to that of Dimensity 8350 or Snapdragon 8s Gen3, which is not very high.

Specifically, Tensor G5's CPU score can exceed 400,000, which has reached the level of Snapdragon 8 Gen3 and Dimensity 9300+. It is also not inferior to Samsung's Exynos 2500. In other words, if we only look at the CPU part, Tensor G5 doesn't seem that bad.

However, in contrast, the "IMG DXT-48-1536" GPU equipped in Tensor G5 has a benchmark score of just over 360,000, which is incredibly low. You know, this GPU score is not only lower than the previous Samsung-designed Tensor G4 (over 440,000), but is even only equivalent to the level of Snapdragon 8 Gen1 several years ago, which is an extremely obvious "weak point."

By now, we can basically conclude that the poor GPU performance is the most obvious reason for Google's self-developed SoC's "failure" this time. So, what exactly is wrong with this GPU?

First of all, the "DXT-48-1536" used by Google is a rather delicate mid-range model in the DXT GPU product series. The "48" in its name means that it can process 48 pixels per clock cycle, and "1536" refers to the GPU's SP (stream processor) scale of 1536.

However, by checking Imagination's official website, it's not hard to find that the top-of-the-line model in the DXT series is the "DXT-72-2304 RT3." Compared with the model used by Google, this "flagship model" has 50% higher single-clock cycle performance and also adds hardware ray-tracing capabilities.

In other words, if Google had been "willing" to use Imagination's best mobile GPU solution, Tensor G5's AnTuTu GPU sub-score might have reached about 1.5 times the current score, that is, close to 500,000 points. This score would be basically at the level of Snapdragon 8 Gen2, thus "matching" Tensor G5's CPU score and making it a "normal sub-flagship SoC" with an AnTuTu overall score of over 2 million.

Actually, there isn't a big problem with Google choosing the relatively "obscure" mobile GPU solution of Imagination for Tensor G5. The problem is that Google was "unwilling" to boost the GPU specifications.

Currently, overseas users are generally very disappointed with the gaming performance of the Pixel 10 series.

So, why was Google "unwilling" to use a higher-spec GPU for Tensor G5? There could be many possible reasons. For example, cost constraints, compromises made considering the SoC's yield rate, or even to make room for other units such as the ISP and NPU in terms of area.

Anyway, Google has finally decided to "self-develop and mass-produce" Tensor G5. This has allowed Imagination's GPU to re-enter the mid - to high - end smartphone market after 9 years (in 2017, Apple announced self - developed mobile GPUs). Considering Google's influence in the European and American markets and the fact that the Pixel series has long had the "exclusive first - release" of new Android systems, even though Tensor G5 doesn't look good in benchmark tests, game and graphics app developers will inevitably have to adapt and optimize their products for the Imagination GPU in Tensor G5 in the future.

Imagination has been doing well in the domestic GPU market in recent years, but its presence in the mobile market has been insufficient.

In this way, the problem of Tensor G5's "weak GPU" becomes less important. Through it, Google has actually achieved multiple goals.

One is to help Imagination prove the "usability" of its mid - to high - end GPUs on today's smartphone platforms, laying a "foundation" for future higher - end products. On the other hand, it's Google's usual "balancing act." By "supporting" Imagination to re - enter the mainstream mobile device market, they can, to some extent, restrict the "dominance" of ARM, Qualcomm, and even NVIDIA (if it enters the market in the future) over the Android GPU ecosystem.

[Some pictures in this article are from the Internet] 

This article is from the WeChat official account "3eLife" (ID: IT - 3eLife), written by 3e Jun, and published by 36Kr with permission.