Huge Google Pixel leaks suggest Tensor upgrades, move to TSMC

Details about Google's upcoming Tensor chips leaked this week, including deeper looks at the Pixel 10's Tensor G5 processor.

First of all: Android Authority published several reports based on details it obtained from documents from Google's gChips division. Initially, the publication reported that Google would abandon Samsung and focus on internal chip development. As part of this shift, the company will shift manufacturing to TSMC. TSMC is currently the leader in this space, making chips for companies like Apple and AMD.

The Tensor G5 (codenamed “Laguna”), which will power next year’s Pixel 10 devices, will likely be the first to be manufactured by TSMC. Android Authority reports that it will use TSMC's 3nm class N3E process – the same one Apple used for the iPhone 16 Pro's A18 Pro and MacBook M4 chips. The Tensor G6 (codenamed “Malibu”) is expected to use TSMC's N3P node, which is also said to apply to Apple's A19 chip.

These changes should lead to efficiency and performance improvements in the Tensor chips, but it remains to be seen how big the improvement will be. Another report from Android Authority looks at the Tensor G5 in more detail and it looks like Google is making some strange decisions.

Minimal hardware improvements

On the CPU side, the Tensor G5 will reportedly continue to use a single Arm Cortex-X4 for the “big” core, which is what we also saw with the Tensor G4 and G3. (CPU is increasingly using a mix of large and small cores to balance performance and efficiency.) Given that the new Cortex-X925 promises significant improvements, it's a shame that Google is sticking with the X4.

However, the company is making major changes elsewhere on the CPU. Android Authority reports that the G5's “medium” CPU cluster will use five Cortex-A725 cores, versus three A720 cores on the G4 and four A715 cores on the G3. The G5 will also feature a smaller “small” cluster with just two Cortex-A520 cores, instead of four A520 cores on the G4.

It's hard to say what impact this will have on performance based on these details alone, but the enlarged mid-cluster could help boost multi-core performance. Ultimately, we'll have to wait until the Tensor G5 lands in the hands of testers.

Elsewhere, the Tensor G5 will reportedly include an Imagination Technologies (IMG) GPU, rather than using the Arm Mali GPU like previous Pixels. Specifically, the chip will have the DXT-48-1536 GPU clocked at 1.1 GHz with support for ray tracing and GPU virtualization. However, the IMG GPU only has two cores. The Mali-G715 in the G4 was clocked at 940 MHz but had seven cores. Ray tracing or GPU virtualization were not supported.

Finally, this brings us to AI hardware. One of the advantages of Google's Tensor chips is that they have custom-made TPUs, which Google says make it possible to deliver AI and machine learning (ML) capabilities on the device that would not be possible with other hardware.

Android Authority reports that the Tensor G5 will feature a faster TPU that can perform up to 18 trillion operations per second (TOPS). However, the company's internal benchmarking shows only a 14 percent speed improvement despite a 40 percent increase in TOPS.

Another interesting note is that the new TPU has small, embedded RISC-V cores that allow the execution of operations that are not implemented in the hardware and support for on-device training.

What does it all mean?

It's always a bit difficult to translate technical details into real benefits for users, especially when we release early leaks and don't have the hardware to actually test them. However, we can find out some details. First, the Tensor G5 will likely have improved performance and efficiency over the G4, although it will likely still lag behind the latest Qualcomm Snapdragon and Apple A-series chips.

However, Google's Pixel phones tend to work quite well even with low-performance hardware, as we've seen with recent Tensor chips. However, as Pixels become more and more expensive, it would be nice if Google made further performance improvements to keep up with the competition.

The G5's TPU improvements could also pave the way for some new AI features, so it will be interesting to see what Google does with the extra power.

Source: Android Authority, (2)