Robert Triggs / Android Authority
Google’s Tensor G5, expected to power next year’s Pixel 10, may be the most anticipated chip in the series yet. For the first time ever, Google is designing a chip entirely in-house, without relying on Samsung to do much of the work. Previous Tensor models, and even current Tensor models, have lagged behind other chipset manufacturers, so this is expected to make it even more competitive. But will it actually happen? So far, we haven’t heard anything about the specs of the new chip, but that finally changed today.
Thanks to a massive leak from Google’s gChips division. Android permissions has viewed an authoritative document from the end of Tensor G5’s design that tells everything about Google’s next-generation chip.
Even more bizarre CPU decisions
The Pixel 9’s Tensor G4 featured an upgraded CPU cluster compared to the previous generation. At least in theory, the actual performance is only marginally faster, so it was only 6% faster in the multi-core test. This is because Google removed the mid-cluster core, presumably to reduce power usage.
Tensor G5 will upgrade the CPU cluster again, but probably not in the way you would expect. Google has decided to keep the same single Arm Cortex-X4 primary core. This is an interesting choice considering the new Cortex-X925 promises some big improvements. We also decided to move the core cluster again. The intermediate cluster now has five Cortex-A725 cores instead of the three Cortex-A720 on Tensor G4, and the small cluster has been reduced accordingly to only two Cortex-A520 cores. I have summarized the specifications below.
Tensor G3 (“Zuma”) | Tensor G4 (“Zuma Pro”) | Tensor G5 (“Laguna”) | |
---|---|---|---|
big cluster |
Tensor G3 (“Zuma”)
1x Arm Cortex-X3 |
Tensor G4 (“Zuma Pro”)
1x Arm Cortex-X4 |
Tensor G5 (“Laguna”)
1x Arm Cortex-X4 |
mid cluster |
Tensor G3 (“Zuma”)
4x Arm Cortex-A715 |
Tensor G4 (“Zuma Pro”)
3x Arm Cortex-A720 |
Tensor G5 (“Laguna”)
5x Arm Cortex-A725 |
small cluster |
Tensor G3 (“Zuma”)
4x Arm Cortex-A510 |
Tensor G4 (“Zuma Pro”)
4x Arm Cortex-A520 |
Tensor G5 (“Laguna”)
2x Arm Cortex-A520 |
Obviously, it’s difficult to predict performance based on specs alone, but with changes like this, Tensor G5 should at least deliver decent multi-core performance gains. But it would be a shame to see the same Cortex-X4 used again.
latest GPU
One area that really surprised me about Tensor G5 while reviewing the documentation was the GPU. Not many vendors offer GPU IP. Google has also used Arm Mali in all of its past Tensor chips. Why would that change now? Well, I don’t know, but it happened. Tensor G5 is equipped with a GPU from Imagination Technologies (or IMG), the DXT-48-1536 running at 1.1 GHz.
Unfortunately, not much is known about the new GPU except for two interesting details. Firstly, it will come with support for ray tracing. This is a first for a Google chip that typically omits such “gaming” features. A second, and perhaps equally important, detail is support for GPU virtualization, which allows virtual machines to use accelerated graphics. Google has been working on various virtualization-based features for some time, so it makes sense that this would be included.
The new GPU specifications are summarized in the table below.
Tensor G3 (“Zuma”) | Tensor G4 (“Zuma Pro”) | Tensor G5 (“Laguna”) | |
---|---|---|---|
GPU |
Tensor G3 (“Zuma”)
Arm Mali-G715 (7 cores) |
Tensor G4 (“Zuma Pro”)
Arm Mali-G715 (7 cores) |
Tensor G5 (“Laguna”)
IMG DXT (2 cores) |
frequency |
Tensor G3 (“Zuma”)
890MHz |
Tensor G4 (“Zuma Pro”)
940MHz |
Tensor G5 (“Laguna”)
1100MHz |
ray tracing |
Tensor G3 (“Zuma”)
not supported |
Tensor G4 (“Zuma Pro”)
not supported |
Tensor G5 (“Laguna”)
supported |
GPU virtualization |
Tensor G3 (“Zuma”)
not supported |
Tensor G4 (“Zuma Pro”)
not supported |
Tensor G5 (“Laguna”)
supported |
Minor improvements for AI
C. Scott Brown / Android Authority
Google’s phones have always had great AI features, and AI is clearly one of the main reasons Tensor chips exist. By incorporating a custom-designed TPU, Google is able to deliver a much better experience than using off-the-shelf chips.
Tensor G5 has a reasonably fast TPU compared to Tensor G4 (which itself was exactly the same as Tensor G3). The TOPS (Trillions of Operations Per Second) value is almost 40% higher, which does not translate well to actual performance. According to Google’s internal benchmarks, the new TPU is only 14% faster. The improved TPU also includes several new features for developers at Google, including a smaller embedded RISC-V core that allows you to perform operations not implemented in the hardware, and support for on-device training. I am. The specifications of the new TPU are summarized below.
Tensor G3 (“Zuma”) | Tensor G4 (“Zuma Pro”) | Tensor G5 (“Laguna”) | |
---|---|---|---|
tops |
Tensor G3 (“Zuma”)
13 / 6.5 TOPS (INT8/FP16) |
Tensor G4 (“Zuma Pro”)
13 / 6.5 TOPS (INT8/FP16) |
Tensor G5 (“Laguna”)
18 / 9 Top (INT8/FP16) |
Improved performance compared to previous generation |
Tensor G3 (“Zuma”)
+65% |
Tensor G4 (“Zuma Pro”)
0 |
Tensor G5 (“Laguna”)
+14% |
As we’ve already revealed, Tensor G5 is built on TSMC’s 3 nm class N3E process node, just like Apple’s A18 Pro, for example. Interestingly, the die size is 121 mm^2, whereas the Apple A18 Pro’s chip is only 105 mm^2, making the G5’s chip noticeably larger.
While the specs for Google’s upcoming Tensor G5 aren’t all that impressive on paper, it’s worth remembering that this is still the chip that will power Pixel smartphones. Software is probably more important here than hardware. It will be interesting to see how Google utilizes the new hardware to bring new features to the Pixel 10 series.
What do you think about the leaked Tensor G5 specs?
202 votes