Google’s Tensor mobile chip has garnered a reputation for being a perfectly adequate processor with heat issues, unfortunately, there doesn’t appear to be any substantive fix coming anytime soon according to leaked documents.
The folks over at Android Authority got their hands on internal documents from Google’s gChip engineers that highlight the company’s acknowledgment of its flagship processor underperforming not only on devices but by internal metrics such as price and expectations.
Not only has Google been tracking the poor performance of its first party chips development over four device cycles, but it looks like the company’s plan to address the phenomenon is to cut back to level the general experience over the next two releases.
According to the documents reviewed by AA, Google isn’t rehauling its chip design, but instead its looking to reduce the die area of the chip to lower its manufacturing cost for starters. Google wants to drop 4 percent of its Tensor die area at 121mm2 down to 105mm2 to put it on par with Apple’s A18 Series chip.
As a result of the overall size reduction of the die Google is also looking to drop a DSP (digital signal processor) core to save space as well as cutting the System-Level Cache support down to 4MB from a high of 8MB in previous chips.
Oh, and say goodbye to the ray-tracing you started enjoying this year because next year, it goes the way of the dodo to make a more effecient processor.
The major impetus for reducing the die also lies with making a more cost-efficient processor in the process. The current Tensor development exceeds profitable or flat financial expectations for the company with aiming for a $65 target or less than half the cost of its competitors. Part of Google’s original plan for creating its own processor was to shift the cost of partnering with Qualcomm for parts to its Pixel lineup, to bring that process indoors.
However, the size of the die negates most efforts by the company to make any profits from its chips now. Next year, Google is said to be using TSMC’s manufacturing to trim down its chips and save some money in the process.
Google also has user feedback asking about the Tensor’s poor battery performance.
Battery life was mentioned several time throughtout the docs reviewed by AA and is something users have continually knocked about the experience across several different handsets.
Interestingly enough, Google also doesn’t seem to be looking to improve the CPU by any meaningful standard as the company plans to continue to use the Cortext-X930 core paired with several Cortex-A730 rather than using the Cortext-A5xx core which could be more powerful but less efficient overall.
Qualcomm has already announced that it’s bringing neural processing units (NPUs) to its next mobile processor which will handle on device processing for things like AI, machine learning and other auxiliary CPU tasks, and Google may not be improving the core CPU experience, but it may bring onboard unannounced CPU additions to make up for some of the development deductions.
Following Apple’s umpteenth A-Series chip announcement, the news that Google would be crafting its own mobile processor to close the stack on Pixel features and device CPU sounded great in theory. Unfortunately for Google, its competitors have managed to outperform the company and its chipset goals in less time. Qualcomm’s Snapdragon processors continue to track impressive numbers while Google’s Tensor chip just overheats and throttles.
Google has gotten good at optimizing its software, now it just needs to manage the processor situation with the same sort of results-based gusto.ware, now it just need to manage the processor situation with the same sort of results-based gusto.
