A mere hours after it was revealed that Google would use TSMC's 3nm 'N3E' and 'N3P' processes for the Tensor G5 and Tensor G6, a new report talks about the chipset's specifications that will be a part of next year's Pixel 10 innards. The disappointing aspect of the latest report is that the Tensor G5 will not feature any custom CPU cores like Qualcomm and Apple have adopted, but the GPU will obtain ray tracing and virtualization support, which are features that have already been embraced by competing SoCs.
This year, chipsets like the Dimensity 9400 feature ARM's new Cortex-X925, providing a massive performance and efficiency boost compared to the Cortex-X4. Sadly, Android Authority's Kamila Wojciechowska reports that the Tensor G5 will be considerably behind its rivals as Google will stick with a single Cortex-X4 as part of the CPU cluster. Fortunately, the plus side is that the company will now use five Cortex-A725 cores instead of the four Cortex-A720 present on this year's Tensor G4, along with two low-power Cortex-A520 cores.
One major reason why Google has the liberty to increase the number of Cortex-A725 cores is the improved manufacturing process of the Tensor G5. The 3nm lithography has allowed Qualcomm and MediaTek to incorporate a performance-only core configuration in their chipsets, meaning that Google could obtain some advantages here. The capabilities of the Tensor G5 cannot be predicted by this CPU cluster, but thanks to the increased Cortex-A725 cores, we should witness a sizable multi-core uplift.
As for the GPU, Google is said to ditch ARM's designs in favor of Imagination Technologies, with the new DXT-48-1536 part running at 1.10GHz. What will set this graphics processor apart from the previous-generation ones is that the Tensor G5 will not have ray tracing and GPU virtualization support. In the past, Google opted to skip 'gaming' features, but seeing as how the competition has been rallying behind these technologies, the company has little choice but to adopt the same approach.
As for the Tensor Processing Unit, or TPU, the Tensor G5 is said to have modest gains over the Tensor G4, with Google's internal documents showing that it is 14 percent faster. As for the die size, the Tensor G5 is said to measure 121 mm², making it larger than the A18 Pro. This should give the chipset a few perks, but we will discuss those improvements separately. For now, we feel that Google missed out on an opportunity to add custom CPU cores and GPU to the Tensor G5, but we can keep our fingers crossed for the Tensor G6.