Has Google's Tensor processor project really been good for the Pixel series?

The arrival of the new Google Pixel 9 series also marks the fourth generation of Google’s semi-custom Tensor processor. Unfortunately, the new Tensor G4 has ended up as more of a stopgap model, offering very little in the way of additional performance or new features over last year’s chipset and handsets. The G4 certainly has not been a disaster; we now have SOS calling and a new CPU, but spending time with Google’s latest flagships has made me wonder if Tensor has been worth all the hassle.

Early Google Pixel smartphones brought in silicon from Qualcomm’s Snapdragon series, mostly matching the flagship chips found in other bleeding-edge handsets. It’s worth remembering that brilliant photos and Pixel-exclusive features were around long before Google started building chips. We had Call Screening on the Pixel 3, for instance. The Qualcomm relationship ended with the Pixel 6 series, where Google debuted its Tensor processor, co-developed with and manufactured by Samsung’s semiconductor divisions. On reflection, this decision has undoubtedly had its pros and cons.

We’re familiar with most of the drawbacks by now. Tensor has lagged a generation (or more) behind its rivals in terms of performance, but this has not notably affected most apps, as we’re well past the point of diminishing returns. Gamers have had a tougher time of it; Google’s choice of graphics cores has offered mid-range rather than cutting-edge performance, and the series refuses to adopt ray tracing when all the other flagship chips have had it for years now. The latest Pixel 9 is no different to the Pixel 6 in that regard.

The locked-in relationship with Samsung has also yielded issues outside of Google’s control. A reliance on Samsung’s Exynos modems has resulted in spotty networking performance and battery drain issues, but thankfully, these have gradually been solved. However, even the latest Tensor G4 hasn’t been able to escape Samsung’s manufacturing process, which has run hotter and been less energy efficient than rival TSMC’s equivalent process throughout recent generations. While Qualcomm and MediaTek switched at the first signs of trouble, Google has been unable to. None of Tensor’s transgressions have been deal-breaking, clearly, but Google’s customers have had to muddle through hardware issues that could have been avoided by buying a chip elsewhere.

Now, Google says it’s not particularly interested in topping benchmark charts (easy to say when the Pixel 9 is well off the pace); instead, it claims to want Tensor to power features you can’t get elsewhere. Certainly, Google’s Pixel series offers best-in-class photography software tools and has the biggest mobile AI portfolio in the business, which lends the Pixel 9 a claim to the cutting edge, even without brute performance. Pixel smartphones have, until this year, also undercut their rivals in terms of price. While only Google knows the real margins, we know that the cost of flagship Snapdragon silicon has climbed year on year, while perhaps Tensor has helped Google remain more competitive than it would otherwise have been.

Still, it’s not hard to imagine that the Pixel series may have ended up very similar without Tensor. The Galaxy S24 series, for instance, supports Gemini Nano, enabling the same Gemini chat and some other capabilities as the Pixel 9s. The phone also sports Circle to Search, live translation, and AI-imagine editing tools. Granted, the Galaxy S24 series is not quite as AI-rich as the Pixel 9, and others are much further behind, but that’s probably more of a software development/investment issue than a hardware one. Not forgetting that both are still dependent on the cloud for a number of their AI features, so the chipset doesn’t matter at all in those instances.t

You also don’t have to look far to see smartphones offering excellent photo capture capabilities, whether it’s Xiaomi’s brilliant software bokeh, vivo’s handheld astrophotography, or Samsung’s solid HDR implementation. Sure, Google has some interesting additions, like Add Me and Top Shot, but again, it’s hard to believe these couldn’t run on other platforms (though only Google knows for certain). Even Apple is quickly closing the AI software gap with its new Clean Up image editing tool.

Google isn’t the only one doing AI, it’s supposed forte, and it’s not even clear that its chip is miles out in front when it comes to AI performance. Qualcomm and MediaTek are quick to claim big when it comes to number crunching as well, and at some point, Google is going to need an as yet illusive killer Tensor exclusive to justify the investment. However, consumers are ambivalent about the latest tech buzzword. Google might have all its eggs in the wrong basket.

Despite how it sounds, I’m not down on Google’s Tensor project. The modern Pixel series has been, and continues to be, absolutely brilliant. I’m just not sure how much of that we can attribute to Tensor itself and whether it’s been worth the tradeoffs, especially now the Pixel 9 costs as much as its rivals. My gut says Snapdragon-inside would have been just as capable and maybe even less problem-prone. But perhaps without custom silicon, Google wouldn’t have had the drive to develop so many software features to leverage the hardware. Ultimately, whether Tensor has been good or bad for the Pixel series is a thought experiment without a good answer.

All the above said, Tensor’s best days could still be yet to come. Next year’s Tensor G5 for the Pixel 10 series is anticipated to bring some big changes as Google goes it alone. Although I’ll say that tentatively, as letting go of Samsung’s hand has already caused Google to miss its first deadline, and there’s no guarantee in-house development will be issue-free.

The Tensor G5 will be built at TSMC (possibly on 3nm) rather than Samsung Foundry, paving the way for a more efficient chip with more headroom for extra performance if needed on the inevitable Pixel 10. Ending reliance on Samsung also opens the door for a wider range of components that could propel the user experience forward. We don’t know the specifics yet, but the dubious Exynos modem could make way for Snapdragon, for instance. We might see a more ambitious GPU setup for gamers, but this is all very speculative at this stage. I’m not convinced we’ll see a major departure from Google’s current development ethos that will suddenly propel the Tensor G5 to the top of the benchmark charts, especially if we take Google at its word about not being too bothered about classic performance metrics.

However, we are certain to see next-generation Google-developed components inside the Tensor G5, which could elevate those Pixel-specific benefits and really make the chipset (and next-gen phones) stand apart from the competition. An improved TPU for AI processing is inevitable after this year’s hiatus, and it’ll be fascinating to see how this pairs with Gemini for next-gen on-device workloads. An upgraded Titan security enclave would also be important if we want these tools to play nicely with our most sensitive information. Computational photography, of course, will likely be in line for some upgraded internal components in the ISP and DSP departments, making for even more powerful photo software tools. All the things that make Pixels create could be made even greater with the G5. Here’s hoping.

Looking back over the past four years, Tensor has probably been a success, but I don’t think it’s quite lived up to its full potential. You haven’t missed out on much if you shopped elsewhere, though the Pixel 9 is challenging that even without a meaningfully new chip. With AI now a quickly evolving differentiator in the mobile space and potentially bigger changes coming with the Tensor G5, I’m optimistic that Google’s efforts will pay bigger dividends in the near future. I might have only just laid hands on the Pixel 9, but bring on the Pixel 10 already.