Google Tensor G5 Performance Analysis: Highlights and Highlights

The landing of the Pixel 10 brings with it the new SoC of the house, the Google Tensor G5 manufactured at 3 nm by TSMCIt is a chip called to correct old ghosts (heat, stability) and to make a tangible leap in CPU, although with a small print that we cannot ignore: The graphic part does not shine equally in all tests and in some cases even falls behind the Tensor G4.

If you're interested in numbers, there's plenty of material here. Geekbench, AnTuTu and real-world gaming tests have been drawing a picture with lights and shadows: the CPU is rising strongly and is approaching high-end rivals, the PowerVR D-Series DXT-48-1536 GPU It shows good thermal resistance, but in certain benchmarks it performs below the previous Mali G4. Let's put it in order and compare all the data to understand what's going on.

Architecture and foundations of the Tensor G5

The G5 debuts a modern CPU configuration with 1× Cortex-X4 (3,78 GHz), 5× Cortex-A725 (3,05 GHz) and 2× Cortex-A520 (2,25 GHz)This 1+5+2 distribution is significantly more up-to-date than the G4 and relies on the jump to the node TSMC N3E (3 nm), key to improving efficiency and containing temperatures under sustained loads.

In memory and storage also updates: LPDDR5X (up to 16 GB) and UFS 4.0 to reduce latencies and accelerate loads. In connectivity, the package is completed with WiFi 7 and Bluetooth 5.4, a step forward compared to the G4 with its 4nm process, UFS 3.1, WiFi 6E and Bluetooth 5.3. As if that were not enough, Google has signed with TSMC for future generations (up to Pixel 14), which suggests continuity in this approach to efficiency and stability and in developments such as ironwood.

In the graphic part, the change of third is complete: Goodbye Mali, hello Imagination. The chosen GPU is a PowerVR D-Series DXT-48-1536 around 1.100 MHz, a component that, on paper, is betting on better efficiency and sustained FPS in long sessions, as well as opening the door to advanced image effects and video enhancements thanks to its integration with Google's own ISP and the hardware-accelerated GPU programming.

CPU: the expected jump

In Geekbench 6, the G5's first records on devices like the Pixel 10 Pro XL and the Pixel 10 Pro Fold point to clear progress. Figures have been seen around 2.276 points in single-core y 6.173–6.203 in multicore. Translated into generational improvement, it is equivalent to a +34–35% in multi and until a +17,3% in single compared to the Tensor G4 (depending on the reference taken).

The photo in front of the competition leaves a curious reading: In single-core, the G5 runs practically on par with a Snapdragon 8 Gen 3 on several devices (S24 Ultra, OnePlus 12 or Xiaomi 14 Ultra), while in multi-core is a little behind. In AnTuTu, the CPU section shows 415.848 points, which places it in the orbit of Snapdragon 8s Gen 3, Snapdragon 8 Gen 3 and Dimensity 9300+, with an approximate improvement of 15% compared to the Pixel 9 Pro XL. Wow, that in general computing the CPU finally feels high-end.

GPU: Synthetic Backward Tests and Nuanced Games

Here comes the tricky part. In the Vulkan test of Geekbench 6, the Tensor G5 has been seen hovering around 3.707 points, while the Tensor G4 reached 9.023 pointsThe gap is enormous and, in some rounds, there has been talk that the G4 is up to 323% more powerful than the G5 in these tests. This is not a minor slip; something is going on. There has been speculation about possible GPU optimization issues or a GPU cut to contain costs, but today there is no official explanation and the sensible thing to do is to wait for updates before passing judgment.

More signals in AnTuTu: the GPU score of the Pixel 10 Pro XL with G5 is around 367.206 points, below the Pixel 9 Pro XL (G4) which over 440.000 in the same test. That would speak of a a drop of nearly 20% generation after generation and I would put the G5's graphics level more or less in the league of a Snapdragon 8 Gen1 in that specific metric.

And what do the games say? Well, there are contrasts. Valorant-Mobile It runs at an average of 60 FPS in High quality, and PUBG Mobile also maintains 60 FPS with the “Smooth + Extreme” setting. Whereas in Genshin impact With the “Maximum” preset, the average stays around 29 FPS, And in Wuthering Waves ("The Stormy Waves") the average is around 44 FPS with the highest visual setting. The hardest note comes with Fortnite: With the “Epic” setting, they appear sharp falls and stuttering, with an average of 25 FPS in a 25-minute session. During that game, the terminal l 42,9 ° C, suggesting that there may have been some throttling.

This mix of data leaves a very clear moral: the G5 offers good FPS in optimized titles or with reasonable settings, But chokes on the maximum profiles of some gamesAgain, improved drivers could move the needle, because thermal stability—as we'll see shortly—has indeed improved.

Thermals and sustained performance

One of the great historical complaints of the Tensors was the overheating. With the G5, Reviews agree that the Pixel 10 stays significantly cooler. than previous generations. Reviews like Damien Wilde's (9to5Google) speak of a phone that It does not get hotter than expected in daily use.; the hot moments are practically limited to the initial setup and specific scenarios.

In my own tests with the Pixel 10 Pro XL, it has been observed that the G5 It reaches its working temperature quickly and maintains it without triggering, which translates into better results sustained performance compared to terminals like the Galaxy S25 Ultra. Even compared to recent models like the Galaxy Z Fold 7 or the Nothing Phone (3), the Pixel 10 has been shown cooler overall. In long gaming sessions, it does feel hot, but The SoC stabilizes quickly and does not become uncomfortable. touch.

That said, cases like the session of Fortnite at 42,9 °C show that the limit exists. It is a context where they can come into play optimization of the game itself and maturity of drivers. Since the Pixel 10 just came out, it makes perfect sense to wait. improvements with updates firmware and the gaming ecosystem.

Local AI and product philosophy

It is worth remembering that the Tensor family is not looking for top the raw power rankings. Since its inception, Google prioritizes a optimized experience for your software, with special focus on Generative AI and on-device capabilities. With the move to 3nm and a revised TPU, Google ensures faster responses and greater efficiency in functions such as instant transcription, offline translation, or voice-guided photo/video editing.

The new G5 TPU offers, according to the data handled, up to 40% faster than the G4 in AI functions that are performed on the device. And here, memory counts: with 16 GB of RAM In several models, the system can sustain heavier AI loadsMany brands are drastically increasing RAM to power these scenarios; Apple is the exception —and is often perceived as at the tail end of AI implementation at the system level—although it has its own strategy.

On a day-to-day basis, several media outlets point out that the G5's CPU upgrade—that +34% approximately— It is not always perceived in a dramatic way when opening apps common or move through networks. The interesting thing, though, is that The performance gap between the base model and the Pro is smaller than before, which better unifies the experience within the Pixel 10 range.

Camera, ISP, and Video: More Control from Google

The G5 debuts a ISP completely designed by Google. That gives more control over the HDR, color and night photography, in addition to enabling ambitious video options. The specifications include 8K and 4K captures at 120 FPS, a combination that had not been so present in previous generations. The key is not just “having 8K,” but the synergy between ISP and AI to achieve finer stabilization and natural-looking noise reduction.

Imagination's new GPU helps cement this multimedia building block: better efficiency and stability under prolonged loads allow for complex image effects and demanding video processing. Although synthetic GPU benchmarks don't follow suit, The visual and camera experience could be greatly enhanced. by the set of ISP + TPU + GPU.

Dedicated security with Titan M2

Google maintains its security seal with Titan M2, a discrete coprocessor that works independently of the main SoC. It is based on RISC-V architecture, with its own memory and cryptographic accelerators, and handles sensitive tasks such as check the Boot, prevent rollbacks to vulnerable versions y limit decryption attempts without the correct key. It has been validated under standards such as common criteria, similar to those of SIM or bank cards.

In practice, this translates into more secure mobile payments and robust biometrics, reinforcing the ecosystem that Tensor G5 itself builds. Again, it is a piece consistent with the approach of experience and trust that Google wants to prioritize over the obsession with the “top 1” in benchmarks.

Pixel 10 Pro Fold and the product context

El Pixel 10 Pro Fold It is Google's third foldable and the first to mount the G5. It also marks the arrival of TSMC to the brand's flip and fold, and comes to a market where Huawei and Samsung They have the advantage, with Oppo, Vivo, and Honor squeezing behind. In this context, Google comes in late, yes, but with a chip that better aims at stability and efficiency, two critical values ​​in formats where heat and consumption are even more noticeable.

By numbers, the Fold's results with G5 are identical to those of the XL in CPU —2.276 single, ~6.173 multi— confirm that the jump over the G4 is the most significant since the first Tensor. GPU performance, as we have seen, is still conditioned by the software maturity and can improve with drivers and patches In the next weeks.

Aggregate Benchmarks: Where Does the G5 Stand?

• Geekbench 6 CPU: ~2.276 (single) and 6.173–6.203 (multi). +34–35% improvement in multi and +17,3% in single compared to G4.
• Geekbench 6 Vulkan: ~3.707 points on G5 vs ~9.023 on G4. Large difference; possible role of drivers/hardware.
• AnTuTu total: ~1.173.221 (in the league of Honor 200 Pro and Motorola Edge 60 Pro with 8s Gen 3 and Dimensity 8350).
• AnTuTu CPU: ~415.848 (+15% vs Pixel 9 Pro XL).
• AnTuTu GPU: ~367.206 (below the Pixel 9 Pro XL; drop ~20% gen over gen; near 8 Gen 1 level in that metric).

Cache and Latency: Why They Matter

While Google hasn't focused on raw cache numbers, it's worth remembering why these blocks matter: Larger L1, L2, and L3 caches tend to reduce latency already improve overall system and CPU performance. In modern designs, good sizing and an effective cache policy can make a noticeable difference in mixed loads, the very type of usage where the Pixels excel.

If we add 3 nm, new CPU cluster, LPDDR5X and UFS 4.0, the logical result is a more agile experience when opening heavy apps, browsing large galleries or exporting videos in tools like CapCut. That is, less waiting and less pulling, even when the GPU score in a synthetic test is not dazzling.

The fight with Snapdragon and the role of drivers

The Tensor G5 CPU already competes head to head in single-core with chips like the Snapdragon 8 Gen3, while in multicore still there is one step left. In GPU, current data places it below the latest batch from Qualcomm and MediaTek, and In some tests even behind the G4. Is it definitive? Not necessarily. GPU Drivers and Game Optimization can move the needle in the coming weeks, and there is a history of substantial improvements following updates.

There are rumors that Google may have GPU trimmed to adjust costs, but there is no official confirmation. What is evident is that The company prioritizes AI, stability and temperature about pushing the limits of peak graphics power. In the short term, Google will have to explain the disparity of results and, above all, correct via software wherever possible, because a breach like the Geekbench Vulkan breach it is not normal for a generational leap.

In parallel, the G5 maintains its story: CPU performance at the level of a recent flagship, a battery of improvements in connectivity, camera and security, and much more controlled temperatures than in the previous two generations. For the majority of users, this combination translates into a more rounded mobile phone in everyday life.

Looking at the whole picture, the Tensor G5 leaves a dual flavor: Finally there is CPU muscle and thermal control to compete competently, and the functions of Local AI is more agile. However, the GPU needs work to live up to what is expected of a demanding 3D top-of-the-range model; the clearest clues point to drivers and optimization rather than an insurmountable physical limitation. If Google fine-tunes this aspect, the package could be a much better fit for the fight against Snapdragon and Dimensity.

Related article:

Google Tensor G5 for Pixel 10: all the new features of the new chip

Isaac

Passionate writer about the world of bytes and technology in general. I love sharing my knowledge through writing, and that's what I'll do on this blog, show you all the most interesting things about gadgets, software, hardware, tech trends, and more. My goal is to help you navigate the digital world in a simple and entertaining way.