Google Will Abandon Qualcomm and Build Its Own Smartphone Processors This Year (cnbc.com) 57
Google announced Monday it will build its own smartphone processor, called Google Tensor, that will power its new Pixel 6 and Pixel 6 Pro phones this fall. From a report: It's another example of a company building its own chips to create what it felt wasn't possible with those already on the market. In this case, Google is ditching Qualcomm. The move follows Apple, which is using its own processors in its new computers instead of Intel chips. And like Apple, Google is using an Arm-based architecture. Arm processors are lower power and are used across the industry for mobile devices, from phones to tablets and laptops.
Google Tensor will power new flagship phones that are expected to launch in October. (Google will reveal more details about those phones closer to launch.) That, too, is a strategy shift for Google, which in recent years has focused on affordability in its Pixel devices instead of offering high-end phones. And it shows that Google is again trying to compete directly in the flagship space against Apple and Samsung. The name Google Tensor is a nod to the name of Google's Tensor Processing Unit the company uses for cloud computing. It's a full system on a chip, or SoC, that the company says will offer big improvements to photo and video processing on phones, along with features like voice-to-speech and translation. And it includes a dedicated processor that runs artificial intelligence applications, in addition to a CPU, GPU and image signal processor. It will allow the phone to process more information on the device instead of having to send data to the cloud. Further reading: Google's New Pixel Phones Features a Processor Designed In-House.
Google Tensor will power new flagship phones that are expected to launch in October. (Google will reveal more details about those phones closer to launch.) That, too, is a strategy shift for Google, which in recent years has focused on affordability in its Pixel devices instead of offering high-end phones. And it shows that Google is again trying to compete directly in the flagship space against Apple and Samsung. The name Google Tensor is a nod to the name of Google's Tensor Processing Unit the company uses for cloud computing. It's a full system on a chip, or SoC, that the company says will offer big improvements to photo and video processing on phones, along with features like voice-to-speech and translation. And it includes a dedicated processor that runs artificial intelligence applications, in addition to a CPU, GPU and image signal processor. It will allow the phone to process more information on the device instead of having to send data to the cloud. Further reading: Google's New Pixel Phones Features a Processor Designed In-House.
voice-to-speech huh... (Score:2)
I feel like that's setting the bar pretty low there... We've had that since wax cylinders..
Re: (Score:2)
Did they mean 'voice to text'...? (seriously asking)
Re: (Score:2)
No, they did mean voice to speech. It will translate poorly phrased rant into something a lot more highfalutin.
Following Apples lead (Score:1, Offtopic)
Re: (Score:3)
Re: (Score:2)
It's only worth doing if there is some real benefit. In Apple's case it was mostly marketing, in Google's case... Well they have done an image processing chip before, from the name maybe this one will have some AI acceleration.
I wonder what they will do about the modem though. Do they have an in-house design or are they going to integrate someone else's IP, or have a separate modem chip like many 5G devices do?
Re: (Score:2)
Mostly marketing in what way? Apple's processors are widely lauded for their performance and capabilities compared to the competition.
Re: (Score:1)
Re: (Score:2)
But on real-world tasks, e.g. Antutu, Qualcomm's parts are way ahead.
Please cite what "real world tasks" that says Qualcomm is "way ahead"
For example with web browsers more cores is a massive advantage as they are all heavily multi-threaded and can parallelize tasks like image decoding, page layout, Javascript compilation and execution etc.
On desktop yes; however, I would consider it a major issue if websites required this a great of processing power from a mobile CPU.
Throw in the reality distortion field and all the nonsense about special sauce that somehow makes Apple CPUs better for unspecified and poorly understood reasons and it's a great bit of marketing.
No there are specific and technical reasons [androidauthority.com] why Apple chips are faster. You however have attributed your lack of knowledge to the "reality distortion field." Since you do not know, there must be zero reasons, right?
Re: (Score:1)
Great example of what I'm talking about. A few artificial benchmarks in a review, suddenly it's the best CPU evar. No, it is mediocre for things people actually do, like browsing the web and computational photography.
Show us some benchmarks for those things instead of Geekbench, Apple's favourite artificial benchmark.
Re:Following Apples lead (Score:4, Insightful)
Great example of what I'm talking about. A few artificial benchmarks in a review, suddenly it's the best CPU evar.
No one is saying that but you. That seems like a strawman argument. What has been said is that Apple A series has beaten Qualcomm ones in benchmarks (aka they are faster).
No, it is mediocre for things people actually do, like browsing the web and computational photography.
And what does computational photography have to do with the assertion that Apple ARM chips are faster? Please cite your evidence that it is "mediocre". Oh wait, that is your pure opinion..
Show us some benchmarks for those things instead of Geekbench, Apple's favourite artificial benchmark.
No. You made the assertion that Qualcomm was "way ahead" in real world tests. You have to prove your own assertion. Shifting the burden of proof, much?
Re: Following Apples lead (Score:1)
Re: (Score:2)
Still not as fast as Ryzen though. They need a big cache just to get that far.
It's a common issue with RISC processors. Instruction density is low due to the instructions being simpler than CISC ones, requiring more of them to complete a given task. ARM even created Thumb just to mitigate this in terms of memory usage, which is a big deal for embedded systems and microcontrollers.
Re: (Score:2)
I forgot to add, their M1 desktop CPU is very similar, mid-range performance for top tier price, and works really well in their laptops because the Intel models where heavily thermally limited. But if you actually want the fastest laptop you can get, buy a Ryzen. You might get 15 hours battery life instead of 20 but that's mostly down to Windows, and it's twice a normal workday.
Re: (Score:2)
Indeed. It basically showed that there is no need to stay with AMD64. Pretty impressive.
Re: (Score:2)
In Apple's case it was mostly marketing.
What are you talking about? Apple wanted more customization than it was getting from Samsung; however, both Samsung and Qualcomm sells CPUs to the widest range of customers and models. Those companies cannot really afford to customize many CPUs to specific customers.
With direct control, Apple was able to transition to 64 bit before those companies. Also Apple is able to modify their phone versions for use in other products like AppleTV. So it was not just "for marketing".
Re: (Score:1)
Re: (Score:1)
I give it 6 months (Score:4, Insightful)
Re: (Score:2)
Re: (Score:2)
Re: (Score:2)
Re: (Score:2)
If that makes you slap your forehead so hard you end up with a ruby red palm divot, don't even begin to consider what we've done here to debase +5 insightful, which is how this remark is presently rated.
New addition table:
* 2+2 = +4 funny
* 2+3 = +5 insightful
* 3+3 = space alien technology
I Wonder... Why Make Their Own? (Score:2)
Re: (Score:3)
Cheaper. It's not like qualcomm is bottom of the barrel processors.
Re: (Score:1)
They are for smartwatches.
Re: (Score:3)
actually, it's going to be 30-40% more expensive. this is going to be 1000+
Re: (Score:2)
Re: (Score:2)
Re: (Score:3)
Re: (Score:3)
Re:I Wonder... Why Make Their Own? (Score:5, Insightful)
The trend in business lately has been moving towards Vertical Integration.
Vertical Integration has some real advantages:
1. You can get the Specs you want, and not negotiate with vendors so they can make something that they can sell to your competitors as well.
2. You don't need pay for profit margin for every part manufacturer.
3. Have more direct control of your supply and workforce.
However Horizontal Integration does offer the following
1. A company specialize in making one thing, can often have economy of scale to sell it cheaper.
2. Risk and Failure can be redirected
3. Don't need to manage an other team of specialist.
However over the past few years, supply chain has gotten much more difficult, Trade Tariffs, Country banning trade, Issues with having the right amount of labor... So many companies that are large enough, are going more vertical, if they have the upfront money, and resources to do this.
Re: (Score:2)
Should call it the Google PS/2 (Score:2)
Will there be android features only available to this processor?
Lifespan (Score:3)
Re: (Score:1)
Re: (Score:1)
Re: (Score:1)
Please explain (Score:1)
What kind of programs would run on this dedicated processor? What language would one write programs for it in? Is the CPU of this processor any different than the 2 bit CPU's (well, wider registers, obviously) we wired up in digital logic 101 back in 1995?
I honestly ask, because every time I see the words "Artificial Intelligence", I could replace them with "a simple computer algorithm" and in the end it is exactl
Re:Please explain (Score:5, Insightful)
Generally speaking, it means better support of operands where some of them are low precision.
Generally, 16 bit floating point data mixed in certain operations with 32 bit data.
Basically, precision that is otherwise deemed useless even for the generally forgiving real time graphics use case can be used acceptably in most AI contexts, and thus processors explicitly designed to optimize for this pay off are branded as 'AI accelerators' while other components handle more generally useful data widths.
Most of the time, AI is referring to taking unstructured input and classifying it into more manageable chunks. E.g. you have a picture and then the 'AI' spits out bounding boxes where it identifies things like 'face', 'cat', or 'hot dog'. Similarly, speech-to-text has this unstructured audio file, and the analysis turns it into a more manageable stream of text.
It feels like AI because we never describe precisely how to do it, but feed example data in and a set of useful analysis building blocks and it applies them in various combinations until the output of it's choice of primitives agrees with a real human classification of the same input.
Following Apple's Lead... (Score:4, Insightful)
"It's another example of a company building its own chips to create what it felt wasn't possible with those already on the market. In this case, Google is ditching Qualcomm. The move follows Apple, which is using its own processors in its new computers instead of Intel chips."
Since we're talking about Google's phones, I'm not sure why the submission mentions Apple's computers rather than their phones. Apple's been putting their own chips into their smartphones for a decade, but only recently started doing so with their computers.
Re: (Score:2)
So you're saying that . . . google is the new Microsoft?
That's fine (Score:2)
One question (Score:2)
Re: (Score:2)
Re: (Score:3)
Re: (Score:2)
Yeah, Sun, Apple, MIPS and SGI never had their own fabs, either. During the '90s, most of the Apple and SGI support chips were made in LSI fabs. MIPS CPUs were made by NEC. Sun's SPARC CPUs were mostly made by Fujitsu and TI. Making chips is expensive.
Not optimistic (Score:1)
I worked at Google not too long ago, probably with some of the very people who will be involved in this project, and I'm not optimistic it will be the next Apple A or M series chip. Not because the people involved aren't very smart and good at what they do, but because Google is a horribly dysfunctional company. It's full of middle management types who are only interested in making sure they get their quarterly bonuses and stock options, and are quite happy to sabotage any internal project to achieve that g
Re: (Score:1)
I predict a trend (Score:2)
Google Will Abandon Qualcomm and Build Its Own Smartphone Processors This Year
And later, it will abandon the processors the way it abandoned Google Hangouts. Sometimes the way this company operates baffles the shit out of me. It just doesn't fail fast (not a bad idea), but it fails in a "not so fast and yet kinda expensive" way.
"features like voice-to-speech" ? (Score:2)
That's what TFS said. It's not clear why one would use a processor to convert voice to speech. All speech is voice, but not all voice is speech. And yet, for most practical purposes, voice and speech are the same thing. There's clearly a relationship and yet I find this very confusing. Is there a pressing need to convert voice to speech in an underserved market? How can I invest in this technology?
Lovely (Score:1)