Want to know who wins when we pit the Google Tensor series against Snapdragon 888 in a one-on-one processor showdown?
Google Tensor vs Snapdragon 888 Series: how the Pixel 6 chip fits
Google’s Pixel 6 series has just hit the market and these are the first phones to feature a custom Google SoC called Tensor. Catch the apple? Are you really using the latest and greatest technology?
Google could have bought chipsets from its partner Qualcomm or even bought an Exynos model from its new friends at Samsung, but that wouldn’t have been so fun. Instead, the company worked with Samsung to develop its own chipset that uses a combination of shelf components and some of its internal machine learning (ML) silicon.
The Tensor SoC is a little different from other high-end Android chipsets available from 2021. Of course, we are saving the benchmarks and all performance and battery ratings for our testing. But we already have a lot of information to dive into a paper comparison between Qualcomm’s latest chipsets (and Samsung too, since we’re at it). Let’s see how they compare in theory.
Google Tensor vs. Snapdragon 888 series vs. Exynos 2100
While the next-gen SoCs from Qualcomm and Samsung aren’t far behind, the Google Tensor chip would compete with the current Qualcomm Snapdragon 888 series and flagship Samsung Exynos 2100 chipsets, so we’ll use them as the basis for our comparison.
Given the nature of their relationship, it’s no surprise that Google’s Tensor SoC relies heavily on Samsung technology in its latest Exynos processor. The modem would for example be borrowed from the Exynos 2100. They share the same MaliG78 GPU, but with Google’s SoC with a 20-core version and the Exynos with 14 cores. The similarities are intended to extend to support for similar AV1 multimedia decoding hardware.
We’ll have to see if the graphics performance is ahead or behind by a few frames of the graphics capabilities of the Snapdragon 888, but it should end before the Exynos 2100 of the Pixel 6. However, we expect the tensor processing unit (TPU) chip provide even more competitive artificial intelligence and machine learning capabilities.
Google’s 2 + 2 + 4 processor setup is an odd design choice. It’s worth exploring a bit more, which we’ll get to later, but the highlight is that two powerful CortexX1 processors are supposed to give the Google Tensor SoC more grunt. Older single-threaded CortexA76 cores can make the chip less multitasking. It’s an interesting combination that dates back to the unfortunate setups of Samsung’s Mongoose processor. However, there are unanswered questions about the performance and thermal efficiency of this design that Google has tried to answer. beautiful.
On paper, the Google Tensor processor and Pixel 6 series look very competitive with the Exynos 2100 and Snapdragon 888 series, which can be found in some of the best smartphones of 2021.
Understanding the layout of Google’s Tensor processor
Let’s come to the big question on everyone’s lips: why should Google choose the 2018 Arm CortexA76 processor for a next-gen SoC? The answer lies in a compromise in terms of surface, performance and warmth.
We unearthed a slide (below) from an earlier Arm ad to help illustrate the key points. Okay, the scale of the board isn’t particularly accurate, but the bottom line is that the CortexA76 is smaller and less powerful than the newer CortexA77 and A78 given the same clock speed and manufacturing process. (ISOComparison) 7nm, but Samsung worked with one arm for a while on a 5nm CortexA76. If you want to pay, the CortexA77 is 17% larger than the A76 while the A78 is only 5% smaller than the A77. Consumption between A77 and A78 by 4, the A76 therefore remains the smallest and least powerful option.
The trade-off is that the CortexA76 offers much lower peak performance. Comparing Arm’s numbers, the company saw a 20% microarchitecture gain between the A77 and A76 and an additional 7% in a similar process with the switch to the A78. As a result, multithreaded tasks can be slower on the Pixel 6 than its Snapdragon 888 rivals, although that depends a lot on the exact workload, of course. With two CortexX1 cores for heavy lifting, Google can be confident that its chip will deliver the right combination of maximum performance and efficiency.