Still 28nm, still Cortex A7...Seems like the absolute minimum response to complaints that the 2100 was holding WearOS back. Meanwhile S4 has gone 64 bit, die shots will tell the manufacturing node.
If it is, it is somewhat surprising, but not crazy. I can't imagine they expect to make tons of money off of this, so the cheapest process possible is probably what they're going for.
It's been confirmed elsewhere that it's 28nm. Qualcomm has also stated that the main CPU is essentially unchanged from the 2100. It's clear they did not allocate a lot of budget, otherwise we would see a A35 based chip instead of the same A7 chip from a few years ago.
The a32 is even more efficient than the a7(it's also aarch32 so you get massive speedups for certain instructions). As others sites have noted, the most interesting aspect to this soc is the new low power core.
Yes, clearly smart watches that are still held back by battery capacity need more powerful CPUs chucked at them! 🙄
The Apple Watch using a 64 bit CPU does absolutely zero in terms of performance or battery usage, it’s just an Apple thing, since they want all of their codebase to switch to 64 bit.
Unless you think Qualcomm is exclusively staffed by idiots, there’s a good chance that they know what they’re doing, and prob. have a better grasp of designing an SOC than your average Anandtech poster.
The A7 is more energy efficient than newer ARM cores, and it’s more than fast enough for a Watch. (Heck, they have tablet SOCs running with Quad or Octa core A7 CPUs on them, so saying things somehow not powerful enough for a Watch is ridiculous.)
You could get an SOC using even less power on a newer node, but who’s going to pay for it.
Smartwatches aren’t exactly in hot demand, and the ones that do sell are the lower priced ones.
And good luck making a 200$ device, with a SOC that’s 2-3 times as expensive because it’s on a 14nm node or even worse: A 7nm node.
It's the who's paying and a volume issue, for Qualcomm, absolutely. Cost, cost, cost. The solution they have come up with is pretty smart considering that they stay on the 28nm. If I get 15 hours of simultaneous HRM and GPS which is their claim, it'll be more than sufficient, for my needs at least.
Because A7 is no longer powerful enough. My ZenWatch 2 lags like s*** on its Snapdragon 400 processor, and theres nothing that's meaningfully changed on these processors that suggests it will be different.
The A32 is more efficient than the A7, unless you don't think ARM knows about their own cores.
The Wear 2100 design was three years old, Qualcomm "knows what they're doing" from an investment standpoint, certainly, but you have blinders on if you don't think this was left to languish, as this chip is barely any different apart from the decent low power core addition.
Cute, but you acknowledge the opposite later. Your original claim was - "Yes, clearly smart watches that are still held back by battery capacity need more powerful CPUs chucked at them!"
Then you came around to "You could get an SOC using even less power on a newer node, but who’s going to pay for it."
Nice moving of the goalposts there. Yes, who is going to pay for it is precisely the issue, but the fact remains that this old chip could have benefitted from a newer node, as I said, and ARM thinks the A32 is more efficient.
I've seen that claim, but no numbers are presented. It may be true for Qualcomm's processors. Then again I wonder why does Samsung use smaller nodes in their wareables? They also have great battery life compared to the current Wear devices. This may be influenced by Tizen possibly being more efficient than Wear OS
Any modern manufacturing mode would reduce power consumption of the SoC. And I'd guess that a nice A53 or A55 would be more efficient than an A7 as well. Both are still in-order architectures and have IPC gains that can be translated to better clock/V states. It's rare that an outdated architecture on an outdated manufacturing node is competitive in anything, let alone efficiency.
"Even the fact that the chipset is built on a 28nm fabrication node (which hasn't been cutting edge technology since 2013), comes down to power savings, according to Qualcomm. Since smaller nodes are generally known for being more power efficient (either that, or a lot more powerful), that may seem somewhat contradictory. But Qualcomm told Android Police it chose not to shrink the process node because the passive drain on a 10nm process is significantly higher than 28nm, and a smartwatch is in idle mode the majority of its time in use. Of course, a 28nm fabrication process is no doubt also less expensive than 10nm, so perhaps that played into the decision, though we can't know for sure."
And this makes sense for watches which are predominantly going to be idle.
Most of what a 64 bit CPU would offer would be wider pointers which even Apple isn't really using:
"It’s to be noted that the 64-bit part of the new CPU might not be as clear as one might think, as there’s been some evidence (Credit @KhaosT) that the Watch might be running in ARM64_32 mode, which would be a 32-bit submode of AArch64 that allows for 32-bit pointers while retaining the ISA advantages of AArch64. Such a mode would be most optimal for a low-memory device such as smartwatch."
There are some advantages to ARM-V8 besides more addressable memory, but I imagine Apple more or less wanted the CPU design to be in-house and to have some redundancy with its phone designs first rather than 64 bit for any inherent advantages it would offer watch software.
ILP32 is not a "CPU" mode, it is a compiler model. It's part of LLVM, and Android could use it for Android Wear if they wanted; to pick up the efficiency benefits of the ARMv8 64-bit ISA, while not paying the RAM costs. But of course for that to happen would require all the various parties (Google, QC, watch makers) to get their act together and actually give a damn about shipping a competitive product. They seem utterly uninterested in doing that...
Meanwhile it's simply not true that "no-one buys smart watches". Apple is probably (all we have is estimates, so we can't be absolutely sure) the largest watch company in the world by both units shipped and revenue. In 2017 they were shipping a million a month. Best guess is this is now a million every fortnight. With Series 4, it'll probably soon be a million a week.
So I guess it's now confirmed. Anand Chandrasekher's AnandTech name is V900! Remember these wise words when the iPhone 5S was released: “I think they are doing a marketing gimmick. There’s zero benefit a consumer gets from that”
Yep. Same old Qualcomm investing the absolute minimum in markets they have a monopoly in, and then spinning 3 year old chips as to the benefit of the customer, as V900 is now doing.
Its Qualcomm. The guys who told us that 64-bit on phones was a dumb idea. What do you expect? They were a company with vision five years ago, they're still a company with vision...
As for S4 die shots, we can hope. But the S4 SiP is one TIGHTLY INTEGRATED piece of kit. No-one ripped the S3 or S2/S1 SiPs apart. The iFixIt teardown just showed us the SiP as a whole, and the X-rays weren't much better. TechInsights did a little better, but not much: http://techinsights.com/about-techinsights/overvie...
(and they gave up before trying to peel the DRAM off the CPU underneath).
Maybe this year there's enough interest that some nice internet folks will make a serious effort to try to get inside?
I thought the obsession with round LCDs was holding Wear back. Some of us like square watches(it is why I bought a ZenWatch 2 AFTER the ZenWatch 3 came out).
Why haven't they improved the manufacturing process to something more "current" like 14nm? I understand them not wanting to manufacture a low volume part in the latest fabs where high volume/high value parts are produced, but we are now talking what.... 2010/2011 era tech?
Smartwatches don't have mass market appeal so investments are going to be minimal at best until companies like Qualcomm see a surge in demand. Keep in mind that wrist watches in general aren't a thing anymore with the sub-50 year old population. Even the most popular fitness-only devices that made something of a splash a couple of years ago are generally collecting dust atop bedroom furniture these days and new sales are decidedly slow. Even though Apple is investing heavily in their watch platform, I suspect they're doing it to increase media attention directed at the company in general in order to generate sales for other products rather than for the watches themselves.
There's something wrong with the article layout. The last column (for the 3200) is covered by the Tweets section. Can only see the first 6 characters of each line in that column.
This is a terrible update. It's proof that Google needs to take things into its own hands to enable competitive Android Wear devices. Samsung and Apple both make their own superior chips. QCOMM has no real competition, and is releasing garbage. And Android Wear devices suffer.
Google probably would be best taking over the project, but they don't have the hardware engineering staff and facilities already so spin up would be a lengthy and expensive prospect for what we already know is a very small market. Even if Android Wear has good hardware behind it, chances are it will remain a niche so it would take many years for Alphabet to turn a profit even assuming wearables experience growth. At the moment, they lack a killer app and are thus in a state of decline amid poor sales, weak support from companies, and privacy concerns recently raised over location and biometric tracking.
"Even Apple is selling more old model watches" This is simply not true. As far as we can tell, cumulative units of Series 3 exceeded Series 2 somewhere around Feb 2018.
As for Fitbit sales, seriously? Apple Watch revenues exceeded FitBit in 2017. Volumes were SLIGHTLY less (about 12 million vs about 15 million), meanwhile Apple has gone up in 2018 and FitBit has gone down.
You're like that guy who, not just at release of the iPhone, but even at release of the iPhone 4, keeps insisting that smartPhones are just a fad and the revenge of Blackberry and Nokia is gonna happen any day now.
There's been a lot of press saying the 2100 was holding the Wear ecosystem back, and was the reason that we've been stuck with relatively bulky watches with so-so battery life, particularly since it's 28nm-based.
But if that's the case, why haven't Samsung's watches, which use Exynos chips, had a radical change in form factor? The Galaxy Watch is virtually the same size as the S3 frontier, which is actually larger than the S2.
Other sources say the 3100 is also 28nm-based (which makes sense given the re-use of 4x A7 cores w/Adreno 304) but I'm not sure that's as much of a hindrance as most people claim IF the ultra-low-power processor can take over most of the load. >95% of the time a smartwatch is on my wrist, the processor is doing nothing other than keeping the OS warm and occasional screen updates.
And the ones that are selling, are the lower priced ones.
Until consumers decide to wanting to pay 4-500$ for one (and even Apple has problems there) there isn’t a market for specialized CPUs on expensive nodes.
The Apple Watch certainly seems to pack a lot more processor punch within a much lower total product volume. I'll be curious to see the node used in die scans. 7nm might be too costly for the segment, but even 14nm would be a big leg up from the 28nm fab used here.
The basic problem here is that Smartwatches are still rather useless devices, unless you have a specific usecase for it (Like fitness tracking. And even then, most people are more likely to get a 200$ fitness tracker rather than a 400$ Smartwatch.)
Sure, smart watches are kinda cool, they’re a neat gadget, but it doesn’t do anything that a cellphone doesn’t already do.
And even in the edge scenarios where a watch is more useful than a phone (like holding something in both hands) the drawbacks are still too big and the usefulness too small for people to pay 400$ for one.
We’ve updated our terms. By continuing to use the site and/or by logging into your account, you agree to the Site’s updated Terms of Use and Privacy Policy.
Facebook
Twitter
RSS
43 Comments
Back to Article
tipoo - Friday, September 14, 2018 - link
Still 28nm, still Cortex A7...Seems like the absolute minimum response to complaints that the 2100 was holding WearOS back. Meanwhile S4 has gone 64 bit, die shots will tell the manufacturing node.Drumsticks - Friday, September 14, 2018 - link
Is it still 28nm? The table has a question mark.If it is, it is somewhat surprising, but not crazy. I can't imagine they expect to make tons of money off of this, so the cheapest process possible is probably what they're going for.
hertzsae - Friday, September 14, 2018 - link
It's been confirmed elsewhere that it's 28nm. Qualcomm has also stated that the main CPU is essentially unchanged from the 2100. It's clear they did not allocate a lot of budget, otherwise we would see a A35 based chip instead of the same A7 chip from a few years ago.tuxRoller - Friday, September 14, 2018 - link
The a32 is even more efficient than the a7(it's also aarch32 so you get massive speedups for certain instructions).As others sites have noted, the most interesting aspect to this soc is the new low power core.
Gadgety - Saturday, September 15, 2018 - link
Yes, that table needs to be updated.V900 - Friday, September 14, 2018 - link
Yes, clearly smart watches that are still held back by battery capacity need more powerful CPUs chucked at them! 🙄The Apple Watch using a 64 bit CPU does absolutely zero in terms of performance or battery usage, it’s just an Apple thing, since they want all of their codebase to switch to 64 bit.
tipoo - Friday, September 14, 2018 - link
Ugh, what year is it.More capable CPUs can also reduce power usage if they complete a task with similar peak power, in less time, and get to sleep faster.
And shrinking to 14, leave alone 7nm, would increase the processors efficiency over the old 28nm node. Like, where have you been, that's not news.
V900 - Friday, September 14, 2018 - link
It’s the year where I’m right and you’re wrong.Unless you think Qualcomm is exclusively staffed by idiots, there’s a good chance that they know what they’re doing, and prob. have a better grasp of designing an SOC than your average Anandtech poster.
The A7 is more energy efficient than newer ARM cores, and it’s more than fast enough for a Watch. (Heck, they have tablet SOCs running with Quad or Octa core A7 CPUs on them, so saying things somehow not powerful enough for a Watch is ridiculous.)
You could get an SOC using even less power on a newer node, but who’s going to pay for it.
Smartwatches aren’t exactly in hot demand, and the ones that do sell are the lower priced ones.
And good luck making a 200$ device, with a SOC that’s 2-3 times as expensive because it’s on a 14nm node or even worse: A 7nm node.
Gadgety - Saturday, September 15, 2018 - link
It's the who's paying and a volume issue, for Qualcomm, absolutely. Cost, cost, cost. The solution they have come up with is pretty smart considering that they stay on the 28nm. If I get 15 hours of simultaneous HRM and GPS which is their claim, it'll be more than sufficient, for my needs at least.WPX00 - Saturday, September 15, 2018 - link
Because A7 is no longer powerful enough. My ZenWatch 2 lags like s*** on its Snapdragon 400 processor, and theres nothing that's meaningfully changed on these processors that suggests it will be different.tipoo - Saturday, September 15, 2018 - link
The A32 is more efficient than the A7, unless you don't think ARM knows about their own cores.The Wear 2100 design was three years old, Qualcomm "knows what they're doing" from an investment standpoint, certainly, but you have blinders on if you don't think this was left to languish, as this chip is barely any different apart from the decent low power core addition.
tipoo - Saturday, September 15, 2018 - link
"It’s the year where I’m right and you’re wrong."Cute, but you acknowledge the opposite later. Your original claim was -
"Yes, clearly smart watches that are still held back by battery capacity need more powerful CPUs chucked at them!"
Then you came around to
"You could get an SOC using even less power on a newer node, but who’s going to pay for it."
Nice moving of the goalposts there. Yes, who is going to pay for it is precisely the issue, but the fact remains that this old chip could have benefitted from a newer node, as I said, and ARM thinks the A32 is more efficient.
tuxRoller - Friday, September 14, 2018 - link
Qualcomm has claimed that the smaller nodes have sufficiently worse static leakage such that it dominates the typical usage pattern.Gadgety - Saturday, September 15, 2018 - link
I've seen that claim, but no numbers are presented. It may be true for Qualcomm's processors. Then again I wonder why does Samsung use smaller nodes in their wareables? They also have great battery life compared to the current Wear devices. This may be influenced by Tizen possibly being more efficient than Wear OStipoo - Saturday, September 15, 2018 - link
I remember that around the 20nm half node, but 14nm certainly seemed like a win on all fronts, let alone 7.Death666Angel - Friday, September 14, 2018 - link
Any modern manufacturing mode would reduce power consumption of the SoC. And I'd guess that a nice A53 or A55 would be more efficient than an A7 as well. Both are still in-order architectures and have IPC gains that can be translated to better clock/V states. It's rare that an outdated architecture on an outdated manufacturing node is competitive in anything, let alone efficiency.V900 - Friday, September 14, 2018 - link
And that’s where you’d be wrong.The Cortex A7 is more energy efficient than the A53 by quite a margin if you’re talking about a device as starved power as a watch.
Look at this test from Anandtech for example: https://www.anandtech.com/show/8718/the-samsung-ga...
Raqia - Friday, September 14, 2018 - link
Leakage current of whatever 28nm process Qualcomm is using is better than smaller nodes:https://www.androidpolice.com/2018/09/10/qualcomm-...
"Even the fact that the chipset is built on a 28nm fabrication node (which hasn't been cutting edge technology since 2013), comes down to power savings, according to Qualcomm. Since smaller nodes are generally known for being more power efficient (either that, or a lot more powerful), that may seem somewhat contradictory. But Qualcomm told Android Police it chose not to shrink the process node because the passive drain on a 10nm process is significantly higher than 28nm, and a smartwatch is in idle mode the majority of its time in use. Of course, a 28nm fabrication process is no doubt also less expensive than 10nm, so perhaps that played into the decision, though we can't know for sure."
And this makes sense for watches which are predominantly going to be idle.
Most of what a 64 bit CPU would offer would be wider pointers which even Apple isn't really using:
https://www.anandtech.com/show/13364/apple-announc...
"It’s to be noted that the 64-bit part of the new CPU might not be as clear as one might think, as there’s been some evidence (Credit @KhaosT) that the Watch might be running in ARM64_32 mode, which would be a 32-bit submode of AArch64 that allows for 32-bit pointers while retaining the ISA advantages of AArch64. Such a mode would be most optimal for a low-memory device such as smartwatch."
There are some advantages to ARM-V8 besides more addressable memory, but I imagine Apple more or less wanted the CPU design to be in-house and to have some redundancy with its phone designs first rather than 64 bit for any inherent advantages it would offer watch software.
name99 - Friday, September 14, 2018 - link
ILP32 is not a "CPU" mode, it is a compiler model. It's part of LLVM, and Android could use it for Android Wear if they wanted; to pick up the efficiency benefits of the ARMv8 64-bit ISA, while not paying the RAM costs.But of course for that to happen would require all the various parties (Google, QC, watch makers) to get their act together and actually give a damn about shipping a competitive product. They seem utterly uninterested in doing that...
Meanwhile it's simply not true that "no-one buys smart watches". Apple is probably (all we have is estimates, so we can't be absolutely sure) the largest watch company in the world by both units shipped and revenue. In 2017 they were shipping a million a month. Best guess is this is now a million every fortnight. With Series 4, it'll probably soon be a million a week.
name99 - Friday, September 14, 2018 - link
So I guess it's now confirmed. Anand Chandrasekher's AnandTech name is V900!Remember these wise words when the iPhone 5S was released:
“I think they are doing a marketing gimmick. There’s zero benefit a consumer gets from that”
tipoo - Saturday, September 15, 2018 - link
Yep. Same old Qualcomm investing the absolute minimum in markets they have a monopoly in, and then spinning 3 year old chips as to the benefit of the customer, as V900 is now doing.name99 - Friday, September 14, 2018 - link
Its Qualcomm. The guys who told us that 64-bit on phones was a dumb idea. What do you expect?They were a company with vision five years ago, they're still a company with vision...
As for S4 die shots, we can hope. But the S4 SiP is one TIGHTLY INTEGRATED piece of kit. No-one ripped the S3 or S2/S1 SiPs apart. The iFixIt teardown just showed us the SiP as a whole, and the X-rays weren't much better. TechInsights did a little better, but not much:
http://techinsights.com/about-techinsights/overvie...
(and they gave up before trying to peel the DRAM off the CPU underneath).
Maybe this year there's enough interest that some nice internet folks will make a serious effort to try to get inside?
Lord of the Bored - Sunday, September 16, 2018 - link
I thought the obsession with round LCDs was holding Wear back. Some of us like square watches(it is why I bought a ZenWatch 2 AFTER the ZenWatch 3 came out).James5mith - Friday, September 14, 2018 - link
Why haven't they improved the manufacturing process to something more "current" like 14nm? I understand them not wanting to manufacture a low volume part in the latest fabs where high volume/high value parts are produced, but we are now talking what.... 2010/2011 era tech?tipoo - Friday, September 14, 2018 - link
It seems to me like they did the absolute minimum to call this an update, after the complaints of the 2100 holding back WearOS.PeachNCream - Friday, September 14, 2018 - link
Smartwatches don't have mass market appeal so investments are going to be minimal at best until companies like Qualcomm see a surge in demand. Keep in mind that wrist watches in general aren't a thing anymore with the sub-50 year old population. Even the most popular fitness-only devices that made something of a splash a couple of years ago are generally collecting dust atop bedroom furniture these days and new sales are decidedly slow. Even though Apple is investing heavily in their watch platform, I suspect they're doing it to increase media attention directed at the company in general in order to generate sales for other products rather than for the watches themselves.phoenix_rizzen - Friday, September 14, 2018 - link
There's something wrong with the article layout. The last column (for the 3200) is covered by the Tweets section. Can only see the first 6 characters of each line in that column.Ryan Smith - Friday, September 14, 2018 - link
How about now?Marlin1975 - Friday, September 14, 2018 - link
Yea the last part is cut off in IE as well.phoenix_rizzen - Friday, September 14, 2018 - link
It works in Chrome on Android, scrolling the page to view the whole table.Won't be near a desktop until Monday, though.
hertzsae - Friday, September 14, 2018 - link
Same for me using Chome, but it looks fine in Firefox.syxbit - Friday, September 14, 2018 - link
This is a terrible update. It's proof that Google needs to take things into its own hands to enable competitive Android Wear devices. Samsung and Apple both make their own superior chips. QCOMM has no real competition, and is releasing garbage. And Android Wear devices suffer.PeachNCream - Friday, September 14, 2018 - link
Google probably would be best taking over the project, but they don't have the hardware engineering staff and facilities already so spin up would be a lengthy and expensive prospect for what we already know is a very small market. Even if Android Wear has good hardware behind it, chances are it will remain a niche so it would take many years for Alphabet to turn a profit even assuming wearables experience growth. At the moment, they lack a killer app and are thus in a state of decline amid poor sales, weak support from companies, and privacy concerns recently raised over location and biometric tracking.V900 - Friday, September 14, 2018 - link
The problem is that the market isn’t big enough for huge investments in new cores and lower nodes.The only Smartwatches that sell really well are the cheap ones.
Even Apple is selling more old model watches than the 3-400$ Generation 3.
Heck, even the Fitbit is outselling Apple Watch Gen 3.
name99 - Friday, September 14, 2018 - link
"Even Apple is selling more old model watches"This is simply not true. As far as we can tell, cumulative units of Series 3 exceeded Series 2 somewhere around Feb 2018.
As for Fitbit sales, seriously? Apple Watch revenues exceeded FitBit in 2017. Volumes were SLIGHTLY less (about 12 million vs about 15 million), meanwhile Apple has gone up in 2018 and FitBit has gone down.
You're like that guy who, not just at release of the iPhone, but even at release of the iPhone 4, keeps insisting that smartPhones are just a fad and the revenge of Blackberry and Nokia is gonna happen any day now.
dudedud - Friday, September 14, 2018 - link
This should be called 2150sing_electric - Friday, September 14, 2018 - link
There's been a lot of press saying the 2100 was holding the Wear ecosystem back, and was the reason that we've been stuck with relatively bulky watches with so-so battery life, particularly since it's 28nm-based.But if that's the case, why haven't Samsung's watches, which use Exynos chips, had a radical change in form factor? The Galaxy Watch is virtually the same size as the S3 frontier, which is actually larger than the S2.
Other sources say the 3100 is also 28nm-based (which makes sense given the re-use of 4x A7 cores w/Adreno 304) but I'm not sure that's as much of a hindrance as most people claim IF the ultra-low-power processor can take over most of the load. >95% of the time a smartwatch is on my wrist, the processor is doing nothing other than keeping the OS warm and occasional screen updates.
V900 - Friday, September 14, 2018 - link
Smartwatches hasn’t taken off.And the ones that are selling, are the lower priced ones.
Until consumers decide to wanting to pay 4-500$ for one (and even Apple has problems there) there isn’t a market for specialized CPUs on expensive nodes.
tipoo - Saturday, September 15, 2018 - link
The Apple Watch certainly seems to pack a lot more processor punch within a much lower total product volume. I'll be curious to see the node used in die scans. 7nm might be too costly for the segment, but even 14nm would be a big leg up from the 28nm fab used here.V900 - Friday, September 14, 2018 - link
The basic problem here is that Smartwatches are still rather useless devices, unless you have a specific usecase for it (Like fitness tracking. And even then, most people are more likely to get a 200$ fitness tracker rather than a 400$ Smartwatch.)Sure, smart watches are kinda cool, they’re a neat gadget, but it doesn’t do anything that a cellphone doesn’t already do.
And even in the edge scenarios where a watch is more useful than a phone (like holding something in both hands) the drawbacks are still too big and the usefulness too small for people to pay 400$ for one.
unrulycow - Friday, September 14, 2018 - link
Can we get this upgraded to something like dual core A55 on a semi-modern process please? This is just patheticV900 - Friday, September 14, 2018 - link
For the same reason you don’t throw a tractor engine into a two person coupe: It’s not the right solution.The Cortex A55 is a tweaked A53. And the Cortex A7 is quite a bit more power efficient than the A53.
As for why they’re not made on newer, smaller nodes: $$
serendip - Friday, September 14, 2018 - link
I've been using a smartwatch powered by an ARM microcontroller for years. Who needs these big stinkin' CPUs?