Definitely a bit of marketing language to try and make the complexities of security understandable. I think of it as similar to a door with multiple, independent, locking mechanisms.
From the Verge article it seems the CPU in SoC would be some licensed spec based on Exynos, so I am guessing we aren't going to see CPU performance close to A series processors, but it does seem like a promising start. Google has an incentive to go all in on CPU design as that helps their data centers and GCP both, so we might see significant gains over time.
I had to buy a new phone and I am really happy that Pixel 6 looks promising, otherwise iPhone would have been the only choice. I really wish Samsung would get its act together on software and updates, as they have really compelling hardware. For now, at least from the preview of Verge it looks like I will probably pick up a Pixel 6.
Apparently they used to be quite different problems, but with today's focus on "performance per watt-hr" as the most important metric, they've merged together. GCP's new AMD CPUs are popular because they're x86 but do more per watt. And all the big cloud platforms are looking into ARM64 chips for the same reason. Arm architecture is crucial in phones. Voila.
At least my understanding was that the SoC would be different but the CPU core of phone SoC itself would be similar with the laptop/desktop/server die. The server SoC would probably feature more cores, with more memory interfaces and higher clock speed due to more thermal allowance.
At least that's what Apple seems to have done between M1 and A series processors.
I'd imagine that if Google were to ever start deploying ARM servers, it could be a big boon. Amazon Web Services and Oracle Cloud already have ARM servers available, and Azure is supposedly working on their own ARM chips, wouldn't be that much of a stretch.
(Disc: I work for Google, but not in Cloud nor Pixel)
It doesn't need to be public facing servers. It could be also for internal uses such as the storage nodes or YouTube transcoding (which has custom decoding/encoding HW as well). I doubt they'd keep the Tensor subsystem for applications like those, though... Except perhaps for YT, where one day you could feed models with frames as they get converted.
>It seems that CPU design for phones and for data centres are quite different problems
Still limited by the same TDP. Just as data point and prospective Apple spend more TDP budget on its mobile CPU core than any current ARM Server CPU core.
> Just as data point and prospective Apple spend more TDP budget on its mobile CPU core than any current ARM Server CPU core.
Are you talking TDP per core? Because there are many ARM server processors with much higher total TDP than Apple's A/M series processors. Even per core, I find it hard to believe that's true for even most ARM server chips, let alone all, given the total TDP of an M1 (10-15W, 4/4 big.LITTLE cores) or an A14 (6W, 2/4 big.LITTLE cores) chip.
Per Core, The Max TDP per core of M1 pushes to 5W. All current ARM server based on N1 has a per core maximum of 3W max. You have specialised core like those on Fujitsu A64FX but they dont push to 5W either with when running with SVE.
i.e What ever core Google decide to work on tensor could easily be used on Server as well. Whether they are the best for it would be a different question though.
> All current ARM server based on N1 has a per core maximum of 3W max.
The N1 based Altra Q80-33 and the Q64-33 both have a per core max over 3W, with the Q64-33 approaching 3.5W/core. Still way under your figure for M1 of course.
I imagine I'm in the minority but I'd rather have a phone a fraction of a millimeter thicker that has a flush back. At this point it's getting out of hand.
I'm the opposite. I'm happy to have a camera bulge if that means better optics and images. I think the idea is that this phone will have the same thickness for most people who use a case.
What's the useful advantage to a flush back? I don't really see any. As long as I can hold the phone, and it can lie on a table or stand without wobbling, it seems to do what it needs to.
It's not like you're stacking phones on top of each other the way you do books or magazines. (And heck if you work in repair or provisioning and actually do stack phones, it's easy enough to turn every other one 180°.)
Also I don't know where you're getting "fraction of a millimeter" from. Camera bumps are significantly more than a whole millimeter. And their function is depth, not volume -- you can't redistribute it across the whole case or anything. You'd have to make the entire phone thicker, and then heavier as well if you're putting anything into the new extra space.
I just think it's nicer honestly. There's no rational thought behind it, it just feels cleaner. I don't really mind extra weight or thickness, it's not like I'm struggling to pick up my phone. I actually have exceptionally large hands so if anything a bigger phone might be better, though I'm not terribly concerned about it.
If that extra millimeter went to extra battery capacity, it would give an extra day of life to the phone. I would appreciate it more than just a flush back.
I had an early Nexus phone (I forget exactly which one) that had a seemingly physics-defying ability to slooooowly (over the course of minutes) slide off apparently-flat-but-actually-slightly-not surfaces, when laid on its back. Not due to alert vibrations, either, but when entirely idle, or even turned off. I think Google accidentally made the back out of some kind of sci-fi ultraslick material or something.
The current generation of smartphones is designed with the expectation that people will use a case. Pixel 6 cases will no doubt account for the camera shape, like all the rest.
I actually hadn't considered that. I don't usually use cases because they make buttons too mushy and take away from the look and feel of the phone and I (so far) don't drop it anyway. I might look for a good thin case or something.
Probably not the minority. I know I would appreciate a flush back but lets be real consumers really don't have a ton of influence on a design choice like this.
Can't phone cases 'mind the gap' though? I suppose they might still leave a little notch across the top of the camera bulge, but I would think that wouldn't bother people too much.
I am hoping to splurge on a phone in 2021, and it's between the Pixel 6 pro and the Fold 3 for me.
As phones they couldn't be more different. The Fold3 aims to completely change the way we experience phones, while compromising on standard phones essentials (camera).
On the other hand, the Pixel6pro looks to be a final evolution of the classical smartphone, maximizing for battery, camera, software-smoothness and hitting all the essentials correctly.
The rumourmill has it that there is a "Pixel Fold" coming this year, featuring Samsung screens. While I'd love a pixel foldable, Samsung has had a few generations to fine-tune their foldable screens, so I don't know how good a first-gen of those can be.
At least this can sit flat on one axis and should stay put when you tap anywhere on the screen, unlike phones with camera bumps in the corner that rock back and forth diagonally
And if I'm looking at it correctly, the lenses will be a bit further (fractions of a millimeter) from the surface when laying like that. I always worry a little bit when laying my current phone down on its back with tiny lens pop-up, how the lens is basically in contact with whatever I set it on.
It seems like this design will make it very difficult to design cases. On most phones with camera bumps, the case ends up sitting flush with the bump and has a hole for it. Here the case would either have to go over the bump (adding thickness) or be attached only along the edge (easy to break)
I never have my phone outside of its case other than dusting - I always have a Spigen of some kind.
I agree with you, but at this point I'm used to cases anyway. If nothing else, I like being able to flip the phone over on its screenside without the screen touching the surface.
Other sources seem to be reporting 6.4" and 6.7" diagonal screens. So I guess even the smaller Pixel is now as large as the iPhone Max?
Guess I shouldn't have gotten my hopes up for a surge of small phones spurred by the iPhone 12 Mini. Which is a real shame, since I'm way overdue for an upgrade but 99% of phones are way too large.
"Tensor"? Google makes some great stuff but I really wish their product naming conventions weren't so confusing.
First they have the Nexus 7, Nexus 4, Nexus 10, and then ... Nexus 5, Nexus 6, Nexus 9, ... oh wait they aren't versions, they are number of inches ... -____-
And then they have Google Hangouts, Google Hangouts Meet, google Meet, Google Chat, Google Hangouts Dialer, Google Duo, Google Uno, uh ....
And then now TensorFlow and then Tensor ... is that like TensorFlow without the flow?
There's also:
- Tensor = the SoC for the Pixel 6 + P6 Pro
- Tensor Processing Unit = co-processor on the Tensor SoC, and as a dedicated chip on other Google Phones,
- Edge Tensor Processing Unit = a dedicated device (USB/M.2/Chip) for edge compute device (R-Pi Style)
I wonder about the baseband (radio - wifi & cell) component. Are they using Qualcomm / Intel standalone chips? Or did they buy an IP core to integrate onto the SoC? I somehow doubt that they did it in-house, as it's a massive effort to match current state of the art.
Unrelated to this article specifically, but the biggest thing I'm looking forward to with the Pixel 6 is the improved updates timeline, as Google has said they will provide updates for at least 5 years, which puts them more in-line with Apple.
Previous Pixels worked great but then inexplicably lost OS updates after only a few years.
>Previous Pixels worked great but then inexplicably lost OS updates after only a few years.
AFAIK it's not inexplicabile, it's just Quallcomm stopping driver support for newer Android versions a few short years down the road, since they're in the business of selling as many chips as possible and can't monetize the users keeping their phone longer like Apple and Google can with their software and services.
Right. My point was that Qualcomm is not doing this themselves. The only snapdragon support in upstream (Linux, Mesa, etc.) is done by dedicated people in the OSS community.
Qualcomm (and MediaTek, etc.) fork the latest LTS kernel or android common kernel, hack it up for that SoC, and dump that on vendors as the msmXXXX kernel.
They'll rebase on top of upstream or android common kernel patches (of that version) for the contracted lifespan, and then pretend that SoC didn't exist.
This seems like such a short-sighted move by Qualcomm, is there some reason that they would do this and risk losing Google as a customer? Or, did they always assume that they would lose Google as a customer and so they thought it wouldn't be worth the investment?
Google really doesn't have much of a choice in the matter. As big as they are, so far it doesn't appear that they've decided that this isn't an important enough problem to make their own phone SoCs. Qualcomm doesn't exactly have many competitors that could be endorsed over them. Huawei/HiSilicon is anathema in US, Samsung uses Qualcomm's Snapdragon over their own Exynos line in the US, Amlogic and MediaTek don't make high-end SoCs, and Apple only makes them for their own phones.
Qualcomm thrives off of the trend of people expecting new phones every year (or every 6 months in some cases). Since they target the mid and high cost devices, all they have to do is make the phone work for a two or so years, and most people buying those phones are going to have gotten bored of their phone by then.
This is one of the main motivations for Project Treble, which is to completely disconnect the phone's userspace from the kernel version it runs on. It's a hard challenge though. Qualcomm and the other vendors for these high turnover products aren't all that interested in supporting it, because they benefit from replacing phones frequently. If you've ever looked at one of the phone kernels, they shove way more into them than upstream would ever accept (the entire radio stack, graphics stack, quick charge logic, etc.). This tends to make the software running on Android phones heavily coupled to the vendor kernel.
The downstream kernels are full of terrible and disgusting hacks. They're just doing the bare minimum to get some semblance of Linux running on the hardware, then forgetting about it almost entirely aside from whatever random security hotfixes they're going to release, and refocusing entirely on their next design. There's no real effort being put into coding for long-term maintainability, that's left to the hobbyist community.
Amortized cost on the update cycle of Pixel was pretty similar to iPhone(at least in some parts of the world). But I am glad that they are doing longer updates.
I just wish batteries become replaceable again, my phone is 3.5 years old and the only thing that is noticeably worse is battery life. If we are gunning for longer update cycles, batteries should ideally be replaceable. I sincerely hope some big country takes up this as an environmentally friendly/right to repair thing and forces all manufacturers to do this.
If you factor in the resale value of a 2 year old iPhone and a 2 year old Pixel, I doubt this is true. And one of the biggest reasons for the high depreciation is the loss of software updates...
I think more than software updates, iPhone has an aspirational value in many parts of the world so there is good demand from people who can't buy a new one. Pixel in itself doesn't hold that value and there are manufacturers who will have an Android phone with Pixel's specs in half the price of Pixel within a year of its launch.
It's not as convenient as a user replaceable battery, but most phones have batteries that are replaceable by repair shows cheaply.
Weirdly my 3.5 year old phone actually has about the same battery life as it started with. As android 3.5 years ago was terrible at battery management, so software updates have compensated for battery degradation.
But all those replacements seem to be with someone pulling apart the phone's components secured with glue. It looks extremely risky and likely to cause damage to phone or components like display ended up getting scratched.
Also, are original spec batteries available easily? I don't want to put something which is not the same as it might cause some battery related accident.
I replaced battery in my smartphone from 2018 last week. Works perfect. It is not that difficult. Look on few guides on YouTube and you'll be fine. Be carefull with battery detatch. Usually it is glued by tape, but isopropyl alcohol helps a lot.
"Tensor"? Google makes some great stuff but I really wish their product naming conventions weren't so confusing.
First they have the Nexus 7, Nexus 4, Nexus 10, and then ... Nexus 5, Nexus 6, Nexus 9, ... oh wait they aren't versions, they are number of inches ... -____- and most people in this world don't even think in inches -_____-
And then they have Google Hangouts, Google Hangouts Meet, google Meet, Google Chat, Google Hangouts Dialer, Google Duo, Google Uno, uh ....
And then now TensorFlow and then Tensor ... is that like TensorFlow without the flow?
- slate form factor < 6"; the phone should fit in my damn pocket
- no hole punch or notch for the camera on the front (how was this even acceptable to consumers, it looks like shit)
- no otherwise irregular protrusions for the back camera or whatever
Is this just impossible now?
Think Galaxy S9, but just with upgraded hardware and maybe smaller bottom bezels.
There are very few things holding me to Android besides familiarity. I'm glad they exist and I think sideloading/3P app stores are important to have "in the world", but the iPhone 12 Mini is just amazing. If it had OLED/high refresh, I wouldn't question my next purchase at all.
How about a phone that is not a phone anymore, but computer in a phone form factor. Because that hardware is that capable for some time. It's so infuriating watching all this effort going into "experience" features that all boil down to shooting selfies and liking posts on facebook.
Never mind "physically building". The whole section about the CPU is pretty evasive about who designed it to begin with. They kind of, sort of, imply it's their own design — "our first custom built SoC", "The team that designed our silicon" — but neither statement actually says that it's their own design.
It's their own design, codenamed "Whitechapel". Evidently they got help from Samsung. As for who will physically make it, either Samsung or TSMC, no one else can fab this 5nm design.
Looks like getting Qualcomm out of the picture is one motivation.
I read the entire blog post and I don't understand the fanfare, it's a custom SoC, presumably to rival Apple's M1. I don't see anything innovative about this, besides heralding "begun, the neural net wars have".
What is the benefit of having all of the ML bits on device? Can models leverage them post training?
> What is the benefit of having all of the ML bits on device? Can models leverage them post training?
Yes, the whole point is to be able do to things like improve personalized speech recognition on-device, image recognition on-device, translation on-device, etc.
Reduced latency. Which has an apparent performance boost versus sending data to Google for processing and waiting for a response.
Potentially improved privacy (this is how Apple tries to sell it). Less data has to leave the phone to gain the utility of the ML models.
Improved device performance. Reduced network use and better specialized chips leading to better performance. Either in terms of better battery life or better time to get the result.
Yes. The whole point is that you have a slow process train the model offline on very large volumes of sample data, then use that trained model to make actual inferences based on data you find in the field. As those models become bigger and more complex, it takes progressively more computing power to run inference on those models. These ML accelerators are effectively the new GPUs — highly specialised processors designed to more efficiently handle highly specialised workloads.
Are you asking what is the advantage of having ML optimized hardware on device? Yes, running inferences is expensive too, especially for video, photo, and speech processing. I would expect this phone to have user noticeable improvements in those three areas.
ah yes, the age-old traditional camera square, first pioneered in the year of 2018. truly a bygone era