The Google Pixel 7 series is here, and with it comes a number of notable improvements across the board. However, one of the most exciting improvements comes in the form of the chipset, as it’s the heart of any smartphone. Google’s Tensor in last year’s Pixel was pretty good, though had a number of issues. This time around, the company has said to improve battery life and other areas in Tensor G2, and the specifications are basically an iteration of Tensor from last year.

Google Tensor G2: Specifications

Google Tensor G2 doubles down on the same CPU formula as Tensor

Comparing both of these chipsets reveals a lot of similarities, with the basics of each set of cores being retained. Google is still keeping its somewhat abnormal 2+2+4 setup and has opted instead to slightly buff the clock speeds of the big and medium cores. In theory, this means that Google will have an advantage in performance in any application that uses two threads for processing, though it depends on how quickly it throttles or not.

Google bumped up the middle cores to a pair of A78 cores as well, which is a very welcome change. It was puzzling last year that Google had opted for a pair of A76 cores given that they are both worse in power efficiency and worse in performance than the A77 and the A78. This should hopefully see some clear performance and efficiency boosts, thus improving battery life as well.

Finally, the same four little cores are kept as on last year’s Tensor. There’s no real change here.

Overall, while it would have been nice to see a complete jump to Armv9 (with a Cortex X2, A710, and A510 trio), that would have required a large redesign on Google’s part, especially given that this is a modification of a design that was already a modified Exynos design. Maybe next year?

Is Google Tensor G2 4nm or 5nm?

While initial reports suggested that Tensor G2 was fabricated on a 4nm process, Google has since confirmed that isn't the case. Google Tensor G2 is fabricated on Samsung's 5nm node. It's unclear if the company is using Samsung's 5LPE or 5LPP, though 5LPP is better than 5LPE and the original Tensor was manufactured on 5LPE. I'd presume it's on 5LPE unless otherwise stated. LPE stands for Low Power Early (and is the company's first-generation 4nm fabrication process), whereas LPP stands for Low Power Plus.

Is Google Tensor G2 faster than the original Google Tensor?

Performance-wise, Google Tensor G2 will perform around the same as the original Google Tensor. The only improvements are in the change from A76 to A78 medium cores, and the clock speeds are increased across the big cores and the medium cores. There are going to be small improvements, but in real-world usage, it's going to feel about the same.

Google Tensor G2’s GPU has a pretty nice upgrade

Google Tensor G2 makes a pretty big upgrade in the GPU department, jumping up from the Mali G78 MP20 to the Mali G710 MP07. The Mali G710 MP07 is a similar GPU to the one that’s in the Dimensity 9000, though there are likely modifications given the “MP07” suffix. For reference, the MediaTek Dimensity 9000 has a Mali G710 MC10. It’s likely that “7” refers to the number of cores (as it does in the case of the original Tensor) but we’ll have to wait and see.

However, the improvements in the Mali G710 come not just from the core count, but the actual structure itself. It promises major improvements not just in graphics, but in Vulkan performance especially. We’ll need to wait and see how much these improvements come to fruition, but in theory, there should be much better performance not just in gaming and other graphics-intensive tasks, but even in emulators like AetherSX2.

Last time around, the 20-core beast of the G78 in Tensor, while powerful, only really retained that power for several seconds before beginning to throttle. It maintained high frequencies to start with, but the high energy usage and heat generated as a result was a major issue. Seeing Google dial it down a bit and lean into a much more normal GPU configuration for mobile does give hope.

The modem is improved, though we're not sure by how much

Part of the reason why the modem in the original Google Pixel 6 series did so poorly was that it was actually a first-generation 5G modem from Samsung -- the Exynos 5123. It had launched as a part of the Exynos 990 SoC in the Galaxy S11 series, so that should give you an idea of how old it was. This time around, Google is using a much newer Exynos 5300 modem.

While there isn't a lot of information about this particular modem (as it just launched and wasn't previously available in any other device), from our own testing it does seem to have improved significantly. Anecdotally, I am now getting signal in places that I couldn't before, including parts of my apartment where it was basically impossible to get any kind of connection.

Google Tensor G2 TPU Improvements

Google is advertising that Tensor G2’s TPU is “next generation”, improving on what was already an impressive addition to the Tensor SoC last year. Last year, Google’s Tensor chipset had references to an “edgeTPU” in drivers. If it’s the same Edge TPU that the company advertises in its cloud platform and in Coral devices, then it’s capable of 4 TOPS at just 2W of power.

In the case of Google Tensor G2, Google says that the TPU is both 60% more powerful and 20% more efficient. Those are some pretty big improvements given that what the Edge TPU was capable of, assuming it was the same, was already pretty good. It powers features like Photo Unblur and Cinematic Blur.

Is Google Tensor G2 just as good as the Snapdragon 8 Gen 1 or the Exynos 2200?

When comparing specifications, Google's Tensor G2 lags behind a lot of the competition. With its unusual core layout and questions surrounding its modem still, this SoC is undoubtedly weaker than the Snapdragon 8 Gen 1 for sure, and by extension, than the Snapdragon 8 Plus Gen 1 too. However, when compared to the Exynos 2200, things get a lot closer. The Exynos 2200 is fabricated on a 4nm process and scores higher on benchmarks, but several users also report a poor experience on Samsung Galaxy S22 devices.

All of this is to say that while Google Tensor G2 is worse on paper, it seems to perform better. That's likely largely thanks to a combination of various software optimizations. Smartphones are more than just the parts that make them, and despite Google Tensor G2 seemingly being underpowered, you likely won't notice it in normal usage... unless you're gaming.

Google Tensor G2 improves the fundamentals

Google Tensor G2 looks like it could be a pretty nice chipset, but we don’t know if it’s enough to improve on where Tensor fell last year. It really doubles down on the best bits of Tensor, while reworking the things that need to be reworked.

However, a chipset is more than just the sum of its parts. There are almost certainly additional improvements and optimizations under-the-hood in drivers that are used to allow the operating system to interface with different aspects of the SoC. These improvements are a lot more intangible, but stack up to give a much-improved experience overall.

From our own testing, the user experience in Tensor G2 is much improved. It doesn't run as hot, it improves where it needs to, and Google overall appears to have been listening. The biggest criticism I would have is not moving over to Armv9 entirely, be it either a switch to one of the following two configurations:

• Cortex X2, A710. A510
• Cortex X3, A715, A510 Refresh

I imagine that we could expect to see something like this in the future if the company continues to modify Arm's core designs.

Google Pixel 7

$400 $599 Save $199

The Pixel 7 packs the second-gen Tensor SoC, a brighter display, and improved cameras.

$400 at Best Buy

Google Pixel 7 Pro

The Pixel 7 Pro is Google's top-of-the-line flagship of the year, featuring the second-gen Tensor SoC, a 120Hz LTPO display, a telephoto sensor, and a bigger battery.

$899 at Best Buy