Google Tensor G6 Chipset Review — Specifications and Benchmarks

Google Tensor G6 is Google's sixth-generation custom chipset for its phones, powering the entire Google Pixel 11 lineup (Pixel 11, Pixel 11 Pro, Pixel 11 Pro XL, and Pixel 11 Pro Fold). This generation brings two fundamental shifts: dropping the Samsung/Exynos modem in favor of a new MediaTek M90 5G modem, and moving to Arm's newest C1-series cores.

CPU

The Tensor G6 uses an unusual seven-core design instead of the typical eight, in a 1+4+2 layout: a single prime Arm C1-Ultra core at 4.11GHz, four C1-Pro cores at 3.38GHz, and two additional C1-Pro cores at 2.65GHz. These cores are from Arm's new Lumex generation, built on the Armv9.3-A architecture. The chip is manufactured on a 3nm process at TSMC (N3P node), following an official correction from Google itself denying earlier reports of a move to 2nm.

GPU and AI

The chip uses an Imagination Technologies PowerVR C-Series CXTP-48-1536 GPU rather than the more common Mali or Adreno, which remains the chip's relative weak point compared to Qualcomm rivals. On the other hand, the chip features a substantially upgraded Tensor Processing Unit (TPU) with 50% more compute power to run the Gemini Nano model and on-device AI features, alongside an advanced imaging pipeline supporting new camera features on the Pixel 11 lineup.

Connectivity and Performance

The chip supports 5G networks via the new MediaTek M90 modem (replacing the Samsung modem used in previous generations), which is expected to improve the heat and battery drain issues that plagued earlier Tensor chips. In performance testing, the chip scored around 1.8 to 2 million points on AnTuTu, and around 6,115 single-core and 2,436 multi-core on Geekbench (with some variance between devices). These results are considerably lower than its direct rival, the Snapdragon 8 Elite Gen 5 (which scores around 3.9 million on AnTuTu), though Google has officially stated it doesn't design Tensor chips to win synthetic benchmarks, focusing instead on heterogeneous computing, AI features, and thermal efficiency.