Tensor 2 benchmark results reveal iteration, not overhaul for Google’s custom Pixel SoC

Google has made it clear to everyone that they won’t only launch an upgrade to the Pixel 6 and Pixel 6 Pro on October 6th, but that these new devices will come with a new version of their in-house Tensor SoC as well. Dubbed Tensor G2, this new chip hasn’t been completely unpacked by Google just yet, leaving us to know very little about it apart from their small statement on the Google Store at this point.

With the next-gen Google Tensor processor, Pixel 7 and Pixel 7 Pro will bring even more helpful, personalized features to photos, videos, security, and speech recognition

via the Google Store

Those are pretty broad statements about a processor and though there have been guesses as to what Google would do with Tensor G2, we’ve not had any confirmation just yet. As these things tend to go, an early tester of the Pixel 7 Pro decided to run a few Geekbench tests and the results are now out on the web for everyone to look at.

Iteration over innovation

From this lengthy thread on Twitter, it looks as if Google is taking a pretty conservative approach with Tensor G2 instead of trying to keep up with flagship processors by companies like Qualcomm. Tensor G2 will come with 2 Cortex-X1 cores, 2 Cortex-A76 cores, and 4 lower-powered Cortex-A55 cores. This looks to be the same basic layout as Tensor G1 with a few tweaks to the overall formula.

Clock speeds look to be cranked up a bit, with the A76 cores pushing up to 2.35GHz and the X1 core getting a bump to 2.85 GHz. The Tensor G2 has also moved to a 4nm process node so we should see a small improvement in speed and battery efficiency due to these factors. If that translates to better thermal efficiency, this could be a very good move from Google in terms of limiting throttling and increasing stability across the board.

The biggest upgrade seems to be in the GPU department, with Tensor G2 getting the Mali-G710 this time around. Arm says this GPU should offer 20% performance bump over the older Mali-G78 in Tensor G1 and 20% power and energy savings, too. Additionally, the claim is a 35% increase in machine learning tasks, and we all know how much weight Google puts into those sorts of tasks on Pixel phones.

According to Wojciechowski’s Twitter thread, Google runs much of their camera pipeline through the GPU, so these changes should benefit the camera experience in pretty substantial ways. Oh, and this change should also mean far better gaming performance, too!

Overall, however, Tensor G2 isn’t a huge leap in SoC performance, but that could turn out to be a win in a few ways for Google. That’s a conversation for another post, but the move to make Tensor G2 a tweaked, better version of the original Tensor chip should help Google keep the Pixel 7 and Pixel 7 Pro performance moving forward while also being a lot more stable this time around. The Pixel 6 and 6 Pro had tons of issues, and keeping with more of a known entity this time around might be the right move for this version of Pixel.

Time will tell, obviously, but I think there’s a good chance Google has made the right call for Tensor G2. Benchmarks only tell a small part of the story after all, so while having a chip that – on paper, at least – falls in-line with those we saw in 2021 sounds like a miss, it could be a solid strategy. If small tweaks make a better chip that is more about stability and battery life, I think it could be a very good move for Google with this year’s hardware. We’ll know more soon enough!