Google Tensor G5

Tensor G5: 3-nm process and “smart” workloads

Tensor G5 is Google’s fifth-generation in-house mobile platform, manufactured for the first time on TSMC’s 3-nm lines. The move to the Taiwanese foundry brings the expected gains in power efficiency and sustained performance, while reducing heat in AI and photo/video processing. According to official statements and independent tests, CPU performance is up by roughly a third, and the tensor accelerator (TPU/NPU) by even more—directly powering on-device features based on Gemini Nano.

Architecture: 1+5+2 CPU cores, new ISP, and an “unusual” GPU

In terms of CPU layout, Tensor G5 uses a 1+5+2 cluster: one “big” core, five performance-oriented “middle” cores, and two efficiency cores. This shifts the emphasis toward a wider “middle” group for steadier multi-threaded performance and less throttling during long tasks.

The key custom block is a completely updated image signal processor (ISP). It enables faster noise reduction, motion deblur, improved skin-tone rendering (Real Tone), and native 10-bit video in the main shooting modes. The ISP+TPU tandem is what drives the biggest leaps in photo/video quality and the speed of Google’s “magic” camera features.

The most debated element is the GPU. Industry coverage indicates Google moved away from Arm’s Mali/Immortalis line to an Imagination PowerVR DXT solution (DXT-48-1536). It’s an atypical choice for Android flagships and comes with quirks in support for certain games/emulators and the lack of hardware ray tracing. Day-to-day apps are unaffected, but pure gaming performance is not G5’s strongest suit.

Memory and storage subsystem

Tensor G5 targets LPDDR5X and modern display/memory controllers. Scores in AI benchmarks and frame-rate stability in games on different Pixel 10 models vary noticeably with RAM capacity and cooling: 16-GB variants with a vapor chamber predictably hold clocks better and score higher than base configurations.

Modem and connectivity: still a “mixed” ecosystem

Despite switching the SoC to TSMC, Google continues to use Samsung’s external Exynos 5400 modem in the Pixel 10 platform. It supports 5G sub-6 and mmWave, FR1+FR2 aggregation with download speeds up to 14.79 Gbit/s (per Samsung’s spec), and NTN features for emergency satellite communication. The trade-off is missing some of the bleeding-edge capabilities of the newest Qualcomm/MediaTek parts, though in real-world networks this is rarely critical.

Performance: synthetics vs. reality

Synthetic tests paint a familiar picture: Tensor G5 is noticeably faster than Tensor G4 but trails the top Qualcomm and Apple chips. In Geekbench 6 the CPU gains ~20–35% depending on device variant, while 3DMark Wild Life Extreme lands around 19 FPS on Pro models—well below Snapdragon 8 Elite phones. In AI tasks, however, the TPU uplift reaches about 60%, which shows up in faster generative features, transcription, and computer vision.

What about day-to-day use? Pixel 10 reviews note that G5 devices run cooler and more stable, wake from sleep faster, suffer fewer multitasking slowdowns, and noticeably speed up “smart” features—from Magic Cue to Camera Coach. If gaming isn’t the priority, the G5 experience feels tuned for comfort.

AI capabilities: focus on Gemini Nano and privacy

Together with an updated Gemini Nano, Tensor G5 executes more workloads on-device: contextual suggestions, summarization, voice features, enhanced live translation, and media processing. Local execution reduces latency and avoids sending sensitive data to the cloud. For the user, that translates into quicker features that you’ll actually use daily.

Photo and video: a “digital lab” on a chip

Google’s bet remains the synergy of ISP+TPU rather than ever-larger sensors. On Tensor G5, night modes, HDR compositing, and stabilization are improved; bursts and video modes are faster; and skin-tone rendering is more natural. Where previous generations sometimes ran into throttling and heat, long recordings and effect processing now behave more consistently.

Who Tensor G5 is for

• Photo/video and AI enthusiasts. You’ll get the most from the new ISP and NPU.

• Those who value stability and battery life. The TSMC 3-nm move delivers tangible gains in thermals and endurance.

• Not for hardcore gaming. In maxed-out titles, competitors still lead—an intentional compromise by Google.

Bottom line

Tensor G5 is the most “Googley” Tensor yet: instead of chasing benchmark crowns, it doubles down on smart scenarios, camera quality, and comfortable everyday speed. The TSMC 3-nm process and refreshed custom blocks (ISP, display controller, TPU) noticeably elevate what people like about Pixel in real life: stability, photo/video quality, and useful on-device AI. If you care most about games and raw charts, Qualcomm/Apple still have an edge. But if everyday experience matters more, G5 is the Tensor upgrade people were waiting for.