Google’s Pixel phones have been powered by semi-custom Tensor processors since 2021, but these chips have never really been a major threat to rival flagship processors in most benchmarks. Now, a Google executive has reportedly claimed that the Tensor G4 in the Pixel 9 series isn’t designed for benchmarks.

“When we are designing the chip, we’re not designing it for speeds and feats. We’re not designing it to beat some specific benchmark that’s out there. We’re designing it to meet our use cases,” Google Pixel product manager Soniya Jobanputra told the Financial Express.

The executive also noted that improving app launch speeds were a priority with the new chip:

We knew that we had a pain point with opening apps. And so as we built G4, we really focused on, okay, what do we need to do to make sure that experience is better for users.

Google Tensor: A history of AI over benchmarks?

This isn’t a new stance from Google, though. Company representatives have long asserted that the Tensor chips weren’t designed with raw horsepower in mind. Google Silicon executive Monika Gupta noted back in 2022 that the company was “comfortable” not winning benchmarks because the Tensor chips were built for AI features.

Fellow Google Silicon executive Phil Carmack also told Ars Technica back in 2021 that they opted for two big CPU cores in the original Tensor chip as it was the best solution for efficiency. He claimed that one big CPU core — seen in rival chips of the era — was only useful for winning single-thread benchmarks. So there’s a clear history of Google dismissing benchmarks as a consideration, subsequently lagging behind rivals in this category.

However, Google’s claim that it prioritizes efficiency and AI use cases does deserve closer scrutiny. For one, owners of Tensor-powered Pixels have complained about battery life for years, although the Pixel 8 series seemed like a step in the right direction. Furthermore, we’ve seen phones with non-Tensor chips offering impressive AI features too, such as the Samsung Galaxy S24 series. So you clearly don’t need a Tensor-powered phone if you want nifty AI use cases.

There’s also evidence that the Tensor G4 in the Pixel 9 series was supposed to be a more ambitious chip. According to previous reports, Google initially wanted the Pixel 9 chip to be a fully custom design manufactured by TSMC instead of long-standing partner Samsung. However, these same reports suggest that Google missed a deadline for this ambitious chip and decided to go for a Tensor G4 chip that was an incremental upgrade over the Tensor G3. These claims seem to bear out as the Tensor G4 shares plenty in common with the Tensor G3 (i.e. GPU, media decode blocks, TPU), with the exception of the CPU and modem.

Opinion post by

Robert Triggs

If you’ve seen our latest Google Pixel 9 exclusive, the phone’s Tensor G4 processor is set to be the smallest change to the series so far. While peak performance has never been a Tensor accolade, there’s little to no upgrade in the chip’s cornerstone AI capabilities either. The Tensor G4 reportedly features exactly the same third-generation TPU, codenamed “rio,” running at the same clock speed as the Tensor G3. The reason is that Google reportedly missed deadlines for a more potent custom chipset, which will now have to wait until the Pixel 10, and had to hastily cobble together an improved Tensor for the Pixel 9 series.

If this holds true, surely the Pixel 8 series should be able to run all of the Pixel 9’s upcoming AI features? Well, the CPU and GPU upgrades appear to be nowhere near big enough to make a meaningful difference to any AI processing, the DSP that runs camera algorithms is the same as last gen, and the identical TPU is the core that binds Google’s on-device AI capabilities together.

I’d argue that the Tensor G4 shares so many core similarities to the G3 that (virtually) the only reason why Google won’t bring its latest features, such as AddMe and Pixel Screenshots, to the Pixel 8 series, at least not in a hurry, is to upsell the Pixel 9. There’s no denying that bringing such features to the Pixel 8 would make Google’s best-ever flagship even more compelling but might undermine launch excitement about the new models, despite the camera and other hardware upgrades. This raises a big question: Just how committed is Google to backporting features via Feature Drops?

Virtually, the only reason Google won't bring the Pixel 9's latest AI features to the Pixel 8 series is to upsell the Pixel 9.

Google’s history with Feature Drops is pretty hit-and-miss. While it has brought plenty of new features to the Pixel lineup over time, we’re still waiting on some of the bigger promises like Zoom Enhance. There’s no guarantee that all of Google’s latest AI features will even be available for the Pixel 9 at launch, so any hope of features making their way back to the Pixel 8 series in a timely manner feels remote. Still, the similarities between the Tensor G4 and G3 make this all the more possible than in previous years, so here’s hoping that, even if there’s a reasonable delay, we see as many Pixel 9 AI features on the Pixel 8 as possible.

There’s one exception to all this — RAM. On-device AI is RAM heavy; it’s the reason the lower-specced Pixel 8 didn’t initially ship with Gemini Nano, while the 8 Pro’s larger 12GB RAM pool made it possible. According to leaks, the Pixel 9 is expected to ship with 12GB of RAM, and the Pro models will all receive a boost to 16GB. That’s a lot more memory than the baseline Pixel 8, but 12GB matches the capabilities of the Pixel 8 Pro. Once again, then, it looks like the more affordable Pixel 8 stands to miss out, but Google’s previous premium model should be capable of matching the Pixel 9. However, just how broken up Google’s AI feature set will become across models remains to be seen.

The Pixel 8's 8GB RAM might be too small, but the 8 Pro is capable of matching the base Pixel 9.

Even so, a processor with few upgrades cuts through the usual inter-generational barrier and puts Google and the Pixel series in an interesting position. Can Google rely on the superb hardware upgrades alone to sell the Pixel 9 while using this opportunity to show that the best software features can transcend generations? We’re already questioning whether seven years of updates really mean the same thing as seven years of cutting-edge features. Google could put the Pixel series on the map as an evolving platform for the industry’s best AI technology, regardless of which generation you buy in. But it’ll have to sacrifice a little Pixel 9 prestige to do so.

Opinion post by

Robert Triggs

If you’ve seen our latest Google Pixel 9 exclusive, the phone’s Tensor G4 processor is set to be the smallest change to the series so far. While peak performance has never been a Tensor accolade, there’s little to no upgrade in the chip’s cornerstone AI capabilities either. The Tensor G4 reportedly features exactly the same third-generation TPU, codenamed “rio,” running at the same clock speed as the Tensor G3. The reason is that Google reportedly missed deadlines for a more potent custom chipset, which will now have to wait until the Pixel 10, and had to hastily cobble together an improved Tensor for the Pixel 9 series.

If this holds true, surely the Pixel 8 series should be able to run all of the Pixel 9’s upcoming AI features? Well, the CPU and GPU upgrades appear to be nowhere near big enough to make a meaningful difference to any AI processing, the DSP that runs camera algorithms is the same as last gen, and the identical TPU is the core that binds Google’s on-device AI capabilities together.

I’d argue that the Tensor G4 shares so many core similarities to the G3 that (virtually) the only reason why Google won’t bring its latest features, such as AddMe and Pixel Screenshots, to the Pixel 8 series, at least not in a hurry, is to upsell the Pixel 9. There’s no denying that bringing such features to the Pixel 8 would make Google’s best-ever flagship even more compelling but might undermine launch excitement about the new models, despite the camera and other hardware upgrades. This raises a big question: Just how committed is Google to backporting features via Feature Drops?

Virtually, the only reason Google won't bring the Pixel 9's latest AI features to the Pixel 8 series is to upsell the Pixel 9.

Google’s history with Feature Drops is pretty hit-and-miss. While it has brought plenty of new features to the Pixel lineup over time, we’re still waiting on some of the bigger promises like Zoom Enhance. There’s no guarantee that all of Google’s latest AI features will even be available for the Pixel 9 at launch, so any hope of features making their way back to the Pixel 8 series in a timely manner feels remote. Still, the similarities between the Tensor G4 and G3 make this all the more possible than in previous years, so here’s hoping that, even if there’s a reasonable delay, we see as many Pixel 9 AI features on the Pixel 8 as possible.

There’s one exception to all this — RAM. On-device AI is RAM heavy; it’s the reason the lower-specced Pixel 8 didn’t initially ship with Gemini Nano, while the 8 Pro’s larger 12GB RAM pool made it possible. According to leaks, the Pixel 9 is expected to ship with 12GB of RAM, and the Pro models will all receive a boost to 16GB. That’s a lot more memory than the baseline Pixel 8, but 12GB matches the capabilities of the Pixel 8 Pro. Once again, then, it looks like the more affordable Pixel 8 stands to miss out, but Google’s previous premium model should be capable of matching the Pixel 9. However, just how broken up Google’s AI feature set will become across models remains to be seen.

The Pixel 8's 8GB RAM might be too small, but the 8 Pro is capable of matching the base Pixel 9.

Even so, a processor with few upgrades cuts through the usual inter-generational barrier and puts Google and the Pixel series in an interesting position. Can Google rely on the superb hardware upgrades alone to sell the Pixel 9 while using this opportunity to show that the best software features can transcend generations? We’re already questioning whether seven years of updates really mean the same thing as seven years of cutting-edge features. Google could put the Pixel series on the map as an evolving platform for the industry’s best AI technology, regardless of which generation you buy in. But it’ll have to sacrifice a little Pixel 9 prestige to do so.

Google’s Tensor processors for its Pixel phones have long been developed in partnership with Samsung and manufactured by Samsung Foundry. We’ve previously heard reports that Google could switch things up with the Tensor G5 inside the Pixel 10 range, and another source has echoed this claim.

Leaker Revegnus posted on X that Tensor chipsets from the Pixel 10 series onwards will be manufactured by TSMC. The tipster also added that Google will use TSMC’s 3nm process (ostensibly for the Tensor G5, at the very least). They also added that Google was expanding its R&D facility in Taiwan and hiring more local semiconductor engineers in preparation for the switch to TSMC.