Google Tensor G6: Why Google Reduced the CPU to Seven Cores

With the Tensor G6, Google has reduced the CPU from eight to seven cores - a rare decision for a new generation flagship SoC. However, this is not a step back: the new Arm C1 cores are faster than their predecessors, the TPU has been significantly upgraded, and the chip is better optimized for local AI and image processing.

Seven Cores Instead of Eight

The CPU of the Tensor G6 is built on a 1 + 4 + 2 scheme: one high-performance Arm C1-Ultra core running at up to 4.11 GHz, four C1-Pro cores up to 3.38 GHz, and two more C1-Pro cores up to 2.65 GHz.

The Tensor G5 had eight cores: Cortex-X4, five Cortex-A725, and two Cortex-A520. The number of cores has decreased, but the new C1 cores are noticeably faster than the Cortex cores used in the Tensor G5.

In Geekbench 6, the new CPU is still faster. The Pixel 11 Pro XL with the Tensor G6 scores around 2677 points in Single-Core and 7321 points in Multi-Core. The Pixel 10 Pro XL with the Tensor G5 shows scores around 2223/5844 points. The difference amounts to approximately 20% in single-threaded and 25% in multi-threaded performance.

However, in terms of CPU speed, the Tensor G6 still lags behind top Snapdragon and Dimensity processors.

Tensor G6 Results in Real Smartphones

The variation is particularly noticeable in Multi-Core: from 6027 to 7321 points. Therefore, tying a single "standard" Geekbench result to the Tensor G6 is meaningless. Cooling, power limits, firmware versions, and specific launch conditions all influence the final results.

No Transition to 2nm

Before the announcement of the Tensor G6, there were speculations about a transition to 2nm, but Google stuck with TSMC’s 3nm process.

The performance boost had to be achieved through other means: via the new Arm C1 cores, higher frequencies, and an updated TPU.

Google also claims up to 20% higher energy efficiency for the CPU, faster application launches, and improved browser performance compared to the Tensor G5.

The TPU Got the Biggest Upgrade

The Tensor G6 features a fifth-generation TPU, which, according to Google, is up to 50% more powerful in machine learning tasks.

This new unit is used for Gemini Nano, photo and video processing, Live Translate, and other functions performed directly on the smartphone.

As a result, some of the performance gains of the Tensor G6 are not well-reflected in conventional CPU benchmarks: a significant portion of computations has been offloaded to the specialized block.

GPU Changes Less Significant

The Tensor G6 uses the graphics processor PowerVR CXTP-48-1536 running at around 1.29 GHz. The Tensor G5 employed the PowerVR DXT-48-1536 at about 1.1 GHz.

The frequency has been increased, but the scale of the GPU update is notably less impressive than that of the CPU and TPU. Therefore, in heavy games, the Tensor G6 still lags behind the graphics of top Snapdragon and Dimensity chips, especially under prolonged load.

For regular tasks, this performance is sufficient for the Pixel, but it does not make the Tensor G6 a gaming flagship.

Titan M3 and MediaTek Modem

The Titan M2 has been replaced with a new security coprocessor, the Titan M3.

Another significant change is the MediaTek M90 modem replacing Samsung, which Google used in previous Tensor models. For smartphones, this is potentially more important than a few percentage points in benchmarking: the modem affects power consumption, heating, and mobile connectivity stability.

The real difference in battery life and heating will be shown in tests of the Pixel 11.

Conclusion

The Tensor G6 has become significantly faster than the Tensor G5, despite the reduction of the CPU from eight to seven cores. The Arm C1 cores have compensated for this decrease, while the fifth-generation TPU has received the claimed boost of up to 50%.

However, the character of Tensor remains unchanged. Google focuses its computational resources primarily on local AI, camera functions, and its proprietary Pixel features, rather than chasing maximum CPU and GPU performance results. The Tensor G6 is notably better than its predecessor, but top Snapdragon and Dimensity processors still lead in pure performance.