The Pixel 10 lineup launching in late 2025 will be the recipient of the Tensor G5, with the latter being Google’s first SoC that will reportedly use TSMC’s 3nm ‘N3E’ architecture. Partial specifications of the chipset, which include a new CPU cluster and ray tracing support, were discussed in a separate report, but sadly, the company will not use in-house cores like Qualcomm has employed in the Snapdragon 8 Elite. So what do all of these changes translate to when talking about the latest benchmark leak? In a word, ‘disappointment,’ as you will find out soon.
The latest benchmark leak shows that Tensor G5 achieves a lower multi-core score than what Apple’s M4 Max obtained in the single-core test
A device called ‘Google Frankel’ was spotted on Geekbench 6 and tested with Android 15 running. As stated before, the Tensor G5 will utilize a new ‘1 + 5 + 2’ CPU cluster, but the chipset will not take advantage of ARM’s newer Cortex-X925 next year but will stick with the Cortex-X4 for some reason, with the five cores belonging to Cortex-A725, and the remaining four being low-power Cortex-A520 cores. Despite the change in the configuration, the Tensor G5 only manages a single-core score of 1,323 and a multi-core result of 4,004.
The M4 Max has a higher single-core score than the Tensor G5’s multi-core figure, which can only mean one of two things. Either the chipset is in its early testing stages, meaning that the scores will obviously be lower than the unit featured in various Pixel 10 models, or Google is least bothered about competing with current-generation silicon. A company executive had previously spoken about the Tensor G4’s lack of impressive performance capabilities, saying that it was not designed to break any records but to improve the user experience.
While we agree that software polishing is equally important to elevate the user experience, the lack of a high single-core or multi-core score generally means that a chip is significantly less powerful than the competition, and that can play a hindrance when running other applications, such as on-device Large Language Models or playing AAA titles. Hopefully, we will witness updated results from the Tensor G5, and the figures will be much higher, so stay tuned.