Igeekphone News, December 15th: Today, technology media Wccftech published a blog post delving into the CPU core architecture of Google’s Tensor G6 chip, revealing that it shares many similarities in design concepts with Mediatek’s Dimensity 9500.
I. Comparison of the Infrastructure between Tensor G5 and Dimensity 9500:
CPU
Dimensity 9500 Tensor G5
The core count is 8 cores
8 cores
One super-large core ARM C1-Ultra with a maximum frequency of 4.21GHz, and one 2MB L2 cache ARM Cortex-X4 with a maximum frequency of 3.78GHz
Three large cores of ARM C1-Premium with a maximum frequency of 3.5GHz, 1MB L2 cache, and five ARM Cortex-A725 with a maximum frequency of 3.05GHz
Four ARM C1-Pro cores (efficiency cores) with a maximum frequency of 2.7GHz, 512kb L2 cache, and two ARM Cortex-A520 cores with a maximum frequency of 2.25GHz
Note: Google has not disclosed the L2 cache of each core
GPU
Dimensity 9500 Tensor G5
Core specifications: ARM Mali-G1 Ultra MC12 Imagination IMG DXT-48-1536
Note: Supports ray tracing and can achieve a 120-frame gaming experience. Does not support ray tracing
Ii. Expected Architecture of Tensor G6:
This chip is expected to make its debut in the second half of 2026 and will be manufactured using TSMC’s 2nm process.
CPU architecture
One ARM Cortex-X930 super-large core
Six ARM Cortex-A730 large cores
One ARM Cortex-A5xx series medium core (efficiency core)
GPU
The 3-core Imagination IMG CXT is even older than the IMG DXT of the Tensor G5
Baseband:
It is expected to abandon the Samsung solution and switch to Mediatek M90, with a maximum downlink rate of up to 12Gbps.
Three. Tensor G6 Borrows two key points from Dimensity 9500:
China National Nuclear Corporation (CNNC) is no longer important:
Previously, there were analyses suggesting that Mediatek enhanced the overall performance of its Dimensity chips by reducing the number of energy-efficient cores. Now, it seems that Google has also adopted this approach.
From the above text, we can see that Google has adjusted the 1+5+2 core layout of Tensor G5 to 1+6+1 of Tensor G6, replacing the efficiency core with an additional large core. With the support of the 2nm process, this change is expected to significantly enhance the overall performance.
Since we are using the ARM public core, let’s use the latest one
One of the key reasons for the significant performance gap between the Tensor G5 and the Dimensity 9500 is that Mediatek uses the latest ARM core, while Google employs a reference core that has been outdated for over two years (Cortex-X4 was released in May 2023).
In contrast, the Cortex-X930 equipped on the Tensor G6 has not yet been officially commercialized. This indicates that Google has finally made up its mind to equip the Tensor with the latest reference core, which is expected to address the long-standing performance shortcomings of the Tensor.
But then again, the GPU of the Tensor G6 is even worse than that of the Tensor G5, which is really intriguing.