Leaked Tensor G2 details checked: Same CPU, everything else gets better

One of the earliest leaks about the Tensor G2 – the new chipset that will power the Pixel 7 series – revealed that the chip will keep the X1, A76 and A55 CPU cores (in a 2+2+4) configuration as the original. Now, developer Kuba Wojciechowski lends more credence to this claim by examining the details of the Geekbench result and other evidence.


Same Cortex-X1 core as before (note: 3396 decimal = 0xD44 hex)

Same Cortex-X1 core as before (note: 3396 decimal = 0xD44 hex)

The Tensor G2 will reportedly be manufactured using Samsung’s 4nm process (the original chip came from Samsung’s 5nm foundries), which allowed for slight bumps in the CPU clock speeds. Two Cortex-X1 cores were capped by 50MHz to 2.85GHz (although this peak could likely be achieved with only one X1 activity), two Cortex-A76 cores got a 100MHz bigger bump to 2.35GHz, based on Geekbench data.

Leaked Tensor G2 details checked: Same CPU, everything else gets better

Yes, the G2 still uses A76 cores, even though they were two generations old at the time the original Tensor was released (they were replaced by the A77 and A78). This does not include the new ARMv9 cores, which have replacements for all three cores – the updated X3, A715 and A510. Moving to ARMv9 required a lot of redesign.

Wojciechowski estimates that clock speed bumps can lead to a 10% (or so) improvement in multi-core performance and hardly any single-core improvement. But the developer also points out that this may not be a bad thing – Geekbench only measures peak performance, not sustained performance. Recent ARM designs have focused on delivering higher performance at a higher heat and power draw cost, which has caused problems for phone makers.

Anyway, the CPU seems to be the only part of the Tensor G2 that hasn’t changed much.

For starters, the Mali-G78 MP20 GPU will be replaced by the Mali-G710 we saw in the Dimensity 9000, for example. ARM claims the new GPU is 20% faster and 20% more efficient than the G78 for rendering graphics and 35% faster for machine learning tasks. The actual optimization will depend on the base count and iteration, so it remains to be determined.

Leaked Tensor G2 details checked: Same CPU, everything else gets better

Although the Tensor chip is named after the Tensor Processing Unit (TPU), the GPU actually does a lot of the heavy lifting in Google’s camera processing pipeline. The TPU itself appears to be part of the second generation, dubbed Jaineiro. The Exynos-based internet service provider may also have been revamped. Overall, the new G2 chip could bring tangible improvements to image processing and AI tasks.

There’s more, as the chipset will reportedly feature Samsung’s new S5300 5G modem, which will be based on the NR Release 16. This promises better performance, stability and efficiency compared to the older version 15 parts.

There’s still some CPU-heavy tasks around, but for the ones Google cares about — the Pixels are known for their cameras after all — the new chip should be a definite improvement.

As for what comes next, Tensor 3 is already in the works and there are rumors that it will be deployed on Samsung’s new 3nm Gate-All-Around process.

source | Across