Google Tensor G5 in the Pixel 10: the highlights and shadows of its performance

  • CPU up: 17–34% jump in single and 35–40% in multi compared to the G4.
  • GPU in question: drops ~20% in AnTuTu and performs worse than G4 in Vulkan.
  • TSMC 3nm and 1+5+2 cluster: better efficiency, thermals and sustain.
  • Enhanced AI and ISP: faster local features and new camera options.

Tensor G5 performance on Pixel 10

new Pixel 10 They are already here and, as expected, the absolute prominence falls on their brain, the Google Tensor G5It's the first Tensor manufactured by TSMC in 3nm and comes with promises of higher performance, improved efficiency, and more controlled temperatures—just what the community has been asking for after several generations of ups and downs in power and stability.

Now, not everything is black and white: the first real measurements bring both good and bad news. In CPU terms, the jump is clear and noticeable on a daily basis, but the GPU has set off the alarms falling behind the Tensor G4 in certain tests. In the following lines, we've compiled and explained all the figures and architectural changes in detail so you can understand what's going on.

Tensor G5 Architecture: TSMC at 3nm and a Modern 1+5+2 Cluster

The G5 marks the abandonment of Samsung Foundry and the jump to TSMC N3E (3 nm), a key change to improve efficiency, thermal stability, and sustained peak performance. Google is opting for an 8-core CPU with a more modern configuration: 1× Cortex‑X4 (up to 3,78 GHz), 5× Cortex‑A725 (3,05 GHz) y 2× Cortex‑A520 (2,25 GHz)This recipe raises the power ceiling with the X4 and reinforces the midrange with five A725s that are very capable in mixed loads.

In the graphic section, Google changes gears and says goodbye to Mali to set up a Imagination PowerVR D-Series DXT-48-1536 GPU @ 1.100 MHzOn paper, it was supposed to improve efficiency and sustained FPS; in practice, the initial results have been mixed, as you'll see below.

The package is completed with LPDDR5X (up to 16 GB), storage UFS 4.0, connectivity WiFi 7 and Bluetooth 5.4 and the security coprocessor Titan M2Compared to the Tensor G4 (4 nm, previous-generation memory and connectivity), it is a clearly more modern SoC with solid foundations for local and multimedia AI.

Tensor G5 Architecture and Specifications

CPU: Significant jump in Geekbench and top-of-the-range performance “from last year”

In CPU, the G5 delivers: in Geekbench 6 results have been seen around 2.296 points in single-core y 6.203 points in multi-core on a Pixel 10 Pro XL, figures that represent an improvement of around 17% in mono y ≈35% in multi compared to data obtained with the Tensor G4 on Pixel 9. In another run (Pixel 10 Pro Fold), the numbers were around 2.276/6.173 (SC/MC), with improvement messages compared to G4 of the 34% in mono y 40% off multi, according to internal comparisons.

Translated into real-life use, this cluster 1 + 5 + 2 It performs like a top-tier flagship in everyday, heavy-duty tasks: opening apps, aggressive multitasking, light photo/video editing, or occasional exports; and if you want optimize Android To squeeze out performance, the G5 offers a solid foundation thanks to the 3nm process.

To give a second reference, in AnTuTu the Pixel 10 Pro XL adds 1.173.221 points, with a CPU block of 415.848That score puts the device on par with Snapdragon 8 Gen 8/3s Gen 8 series handsets and Dimensity 3/8300+, which fits with the “recent high-end” perception in general computing.

Tensor G5 Benchmarks

GPU: Worrying Vulkan numbers and a drop against Tensor G4

The sticking point is in the Imagination DXT‑48‑1536 GPU. In Geekbench 6 (Vulkan test), the G5 scores around 3.707 points, while a G4 Tensor reaches 9.023. We talked about the G4 being up to a 323% more powerful In this scenario, the result is so anomalous that it suggests immature drivers or specific hardware limitations.

The picture is repeated in AnTuTu: the Pixel 10 Pro XL throws some 367.206 points in the graphic section, below the >440.000 that the Pixel 9 Pro XL usually obtains with its Mali-G715 MC7. That is, a ~20% drop in GPU intergenerational in this battery of tests, placing the G5's graphics part in ranges closer to 2021-2022 platforms in purely 3D loads. If your goal is maximum performance in AAA mobile gaming, today the numbers indicate that there are more compelling GPU alternatives within Qualcomm and MediaTek.

There is no official explanation for now. Rumors of a possible performance reduction have leaked. GPU to control costs or consumption, but the reported reduction is too large to be explained solely on this score. The reasonable course of action is to wait for driver/firmware patches and re-measure; even so, graphics performance is the G5's Achilles' heel right now.

AnTuTu in context: good CPU and memory, graphics that do not accompany

With 1,17 million In AnTuTu, the Pixel 10 Pro XL is rubbing shoulders with phones that mount 8s Gen 3 and certain high-end Dimensity. The imbalance comes when looking at the detail: the CPU performs very well and the UFS 4.0/LPDDR5X memory pushes fluidity, but the score of GPU weighs down the totalFor demanding games, this translates to less generous FPS or more conservative graphics settings.

If your profile is comprised of social media, photos, maps, videos, and the occasional game, the G5 will give you a reliable and stable experience. If, on the other hand, you want maximum performance in AAA mobile gaming, today the numbers indicate that there are more compelling GPU alternatives within Qualcomm and MediaTek.

In any case, there are nuances: many experiments play on “mixed terrain” (CPU + NPU + GPU). In these scenarios, the Tensor G5 he defends himself well thanks to its updated CPU and AI acceleration, as we will see in the next section.

Local AI, revised TPU, and “more Google-like” ISP

The hallmark of the Tensor family is its TPU/NPU, with a clear push to prioritize artificial intelligenceWith the jump to 3 nm, Google is talking about improvements of up to 60% in AI loads compared to the previous generation and other internal measurements place accelerations of around 40% in key functions. In practice, this means faster responses in transcription, offline translation and voice-guided editing, and above all the ability to maintain those tasks without falling into throttling after a few minutes.

The camera is the area where the ISP + AI combination is most noticeable. Google launches a Own ISP with more control of HDR, color and night, as well as enabling 8K capture y 4K to 120 FPSThe combination of computational processing and the main sensor produces share-ready images with minimal user intervention, supported by refined noise reduction and stabilization algorithms.

From the hand of that local AI also comes “Camera Coach”, a mode that suggests framing, focal length, and basic composition rules, designed to educate the user's photographic eye. It's a different way of using AI: less "fixing it later" and more "taking a better shot from the start."

Titan M2: Dedicated Security and Verified Boot

Beyond performance, Google maintains its commitment to to maximise security and your enjoyment. to Titan M2, a chip based on RISC-V With its own memory and cryptographic accelerators, it operates independently of the main SoC. Its job is to protect sensitive data by verifying that the system is legitimate upon startup and limiting decryption attempts if authentication fails.

The module has been validated under standards such as common criteria, the same ones used in SIM cards or bank chips. Combined with Android and the Pixel's long update policy, it offers a solid foundation for mobile payments. robust biometrics and mitigation of advanced attack vectors.

Display, audio, and biometrics: how the G5 feels overall

The Pixel 10 Pro maintains a panel LTPO OLED with 1‑120 Hz refresh rate and native resolution of 2.856 × 1.280 (It arrives set to 2.410 × 1.080 to save battery life.) Colors can be adjusted between more natural or more saturated profiles, and brightness reaches ≈2.000 nits in HDR with peaks of up to 3.000 nits in small areas.

Outdoors visibility is good, although some users notice that the auto brightness It falls a bit short and has to be pushed by hand. The glass is prone to fingerprints (on the front), while the back offers a premium feel. The card reader ultrasonic fingerprints on screen it remains fast and well located.

The stereo audio more than meets the volume level, but if you turn it up more than 90% Some distortion may appear. There is support for spatial audio and software enhancements (system-wide automatic captioning and voice isolation on calls) that are useful in noisy environments.

Battery and charging: correct autonomy and conservative times

With 4.870 mAh, the battery life is consistent with that “long day” for varied use with networking, YouTube, music playing in the background, and the camera. It's not a record, but it is consistent, something that the G3's 5nm process contributes to. Where Google remains true to its philosophy is in the load: prioritizes temperature and battery health with a more linear curve that takes about two hours to complete.

In a specific measurement with a PD charger >30 W, the sections were these (approx.), a behavior that puts stability over haste:

  • 10 minutes - 15%
  • 20 minutes - 27%
  • 30 minutes - 37%
  • 40 minutes - 47%
  • 50 minutes - 56%
  • 60 minutes - 65%
  • 70 minutes - 74%
  • 80 minutes - 82%
  • 90 minutes - 90%
  • 100 minutes - 95%
  • 110 minutes - 98%
  • 117 minutes - 100%

Hay wireless charging with Qi2 standard and support for MagSafe-type magnetic accessories. The magnets aren't as strong as on a recent iPhone, but the compatibility is great news if you already have magnetic docks or batteries at home.

Camera: “house brand” processing, brilliant 5x telephoto and an amazing superzoom

The photo module replicates last year's configuration: 50 MP main (1/1,31”, Dual Pixel, OIS), 48 MP ultra wide angle (123rd) and 48 MP 5x telephoto (OIS), with a 42 MP frontThe beauty is in the processing: the G5 produces very well-cooked JPEGs, ready to share, and allows you to shoot in RAW+JPEG without breaking a sweat thanks to his greater muscle.

The main camera is that “point and shoot"Reliable, delivering eye-catching color and dynamic range. At night, detail is preserved and highlights are controlled, while the ultra-wide angle, while good during the day, struggles more with highlights and fine detail in adverse conditions.

El tele 5x It's a delight because it generates a very pleasing natural blur even with somewhat distant subjects. Plus, at full resolution, it can be "stretched" to 10x equivalents quite gracefully. In portraits, the 5x avoids the artifacts of software portrait mode and preserves skin and contours. more natural.

Where there is controversy is in the superzoom (50x-100x). The phone warns that the AI ​​can "invent" content to reconstruct details. On watches with tiny text or very distant objects, these hallucinations are common (changed words, reworked textures). The final photo is striking, but it's important to know that it doesn't always reflect the exact reality.

Another important detail: with the automatic target selection If enabled, in low light, the system may choose to crop from the main sensor instead of using the actual 5x telephoto lens, “simulating” 5x framing. If you want to ensure that the lens shoots at telephoto, disable that option and choose your lens manually.

In video, there is 8K30 and 4K up to 120 FPSThe best results (noise/artifacts) are achieved at 4K30, with solid colors and focus that can be locked onto the subject with a tap. When moving, digital stabilization can be aggressive and generate artifacts; disabling the "Video Optimizer" also removes this digital layer, improving naturalness at the cost of losing some extras.

Additional benchmarks, app availability, and comparison with rivals

tensor g5 benchmarks

A nuance to keep in mind: Geekbench It is not yet officially available for the Pixel 10 series, so some of the results seen come from user units and leaks. There have also been tests on AnTuTu shared by the community (including Reddit), with the total figure of 1,17 million and the CPU/GPU breakdowns already discussed, and the integration with Android 16 in Pixel.

Compared to what is coming, platforms like Snapdragon 8 Elite 2 o Dimensity 9500 They aim to take another step up in graphics and CPU. Google's move is to rely on its ecosystem (Android 16, local AI, computational photography, 7 years of support) to compete "on experience." Even so, the price and positioning of the GPU force explain the proposal very well.

Price, colors and who it is aimed at

The Pixel 10 Pro XL is offered in colors Moonstone, Jade, Porcelain and Obsidian, with a starting price of around €1.299 in Europe ($1.199 in the US). It's a phone for those who prioritize computational photography, local AI features, security, and a polished Android with long-term support, accepting that the pure gaming is not his specialty this year.

As a quick guide: if you're coming from a Pixel 9, the jump in CPU, thermals, and AI features is welcome, but you might notice some steps backwards in demanding titles. If the phone is good for networking, "non-extreme" photo/video, maps, and multitasking, the Tensor G5 offers a more powerful experience. fluid and coherent with what is expected of a modern high-end.

After putting all the pieces together, the reading is clear: the Tensor G5 introduces the much-needed twist to the process (TSMC 3 nm), CPU architecture, and AI/ISP platform, elevating the overall experience and stability; however, the graphics part is still awaiting a response—and hopefully a solution—because it currently measures below the previous generation in several tests. If Google fixes the GPU with drivers, we will be looking at the most well-rounded Tensor; if not, we will have to buy with that small letter in mind.

Google Tensor G5-0
Related article:
Google Tensor G5: The new heart of the Pixel 10 and the jump to TSMC