I think your wrong. The Cortex-A are the custom chips that use the instruction set but aren't a specific design. Just like the Cortex-M are just IP blocks in things like modems and other chips throughout any device.
No, Cortex A is a specific line, and things like Krait, and Apples Swift and Cyclone are NOT Cortex A, even though they are ~compareable. Stuff like the MTK, Allwinner, Exynos, and some of Quallcomm's designs that use the stock ARM IP (Like Snapdragon 610 and 400) are Cortex A. Also, some of the earlier Apple SOC's like the A4 and A5 (which were Cortex A8 and A9, respectively)
No... ARM doesn't make any chips. ARM just sells designs, so while it does include Qualcomm and Samsung, I would strongly lean towards the interpretation that these numbers only reflect ARM silicon shipments. Custom from-scratch designs like Apple's A7 or Qualcomm's Krait probably are not reflected in the data. Both Samsung and Qualcomm have shipped chips that use ARM designs, and both have shipped chips that do not use ARM designs but are ARM compatible. (Remember: Apple does not manufacture chips. Apple gets Samsung to manufacture their chips, traditionally.)
Not really. As the other posting shows, Apple *assembles* the various blocks of ARM architecture into an A7. That's hardly "from scratch". An English major likely couldn't do it, but a competent EE sure could.
Ummm...you are dead wrong. The cyclone core you mention is ARMv8 instruction set compatible, but is by absolutely no means assembled from off the shelf IP blocks. Did Apple license a GPU for inclusion in the A7, yes. Did they license a CPU for inclusion in the A7, no.
With this article I am now interested in seeing if Intel too follows the lisencing of IP route. Going by the way ARM is going, this is what appears to be the logical route, so so I think
Look at the turnover of ARM, look at the turn over of Intel. It's completely impossible for them to do what ARM does. Ironically they exist that way because they could not compete with Intel - instead of turning into AMD mk2 then stayed very lean and mean, living off $0.5 profit/chip. They went for the market Intel didn't want (embedded) and didn't attempt to control the market (like Intel and x86) but licensed designs to everyone.
This is much more appealing to everyone using their chips as it's cheaper and gives them more freedom. Hence Intel's great problems getting into the ARM markets - even if they can produce competitive chips Intel can't survive on the tiny profits ARM makes, and no one wants to go back to being controlled by Intel anyway so will often license ARM in preference to buying x86.
ARM entering the server market with small cores has always been just the starting point, the intention to go bigger was always there.Ofc the industry is slow ,first they'll need to gain some market to develop the software ecosystem and might take some time before we see big cores. On the desktop side i keep saying that i wish the ARM guys would define some new standards and have a go at it.Wouldn't it be nice to start with a clean slate and modernize the desktop? Oh well, maybe AMD will sell us the server chips ,maybe Denver is good ,maybe Steam OS gets some ARM machines, or maybe not.
since 1995? seems reasonable. How many times have you purchased an ARM-based device in the past 20 years? personally I've purchased at least a dozen, and I am just one person.
It makes perfect sense, a common smart phone uses around 10-15 ARM-cores today, and has always used at least 2-3 ARM-cores back to the feature phone days. ARM licenses cores. The processor IP is per unit (so that can be say 10 per chip) not per chip. IP is the keyword here.
"The bigger story was that the 10 billion in 2013 brought the cumulative total for ARM based processors to over 50 billion (note that these are discrete ICs, multiple cores within a single design are not counted multiple times)."
They are not counting individual cores, but discrete IC's. Granted, there may be a couple discrete IC's in each smartphone that ARM is licensing, but it is not an inflated count based on cores.
Latest data I found was 1 billion Android activations as of Sept 2013, with 1.5 million activations/day as of April 2013. So say in the last year they did an extra 750 million device activations (assumes activations/day continues to increase), and maybe another 500 million IOS devices that licensed the specific cores, and you have 2.25 billion "smart" phones. Say they have 3 IC's each on average licensed from ARM (no idea if that is a reasonable number or not, but that's what I'm going with), and you have a total of 6.75 billion chips from smart phones. If that math is within an order of magnitude that means ll the Nokia, Motorola, etc feature phones from the mid-90's on still make up the vast majority of their mobile total. But that would seem to be at odds from Anands statement that "37.5 of the 50 billion chips shipped in the past five years (2009 - 2013)". Either my estimates of smart phones/tablets totals is way off, the amount of licenses ARM gets per unit is significantly underestimated, or there is still a ton of feature phones shipped in the last 5 years.
No longer sure of the overall point of this post, but it's already typed, so I'll submit it.
Data I found (smartphone shippments): 2010: ~300 m 2011: ~500 m 2012: ~650 m 2013: ~ 1 b = 2.45b Add to that some few hundred million tablet shipments, laptops, who knows how many dumb and feature phones....
Read the damn graphs, it clearly says IP shipments! Each ARM core is an IP shipment. If you count IC's then a modern phone basically has a two, three or four ARM-powered chips. In reality these represents over 10 IP licenses from ARM. Just a modern application processor with built modem has at least something like 5-10 ARM cores. Nowadays there is usually a couple of ARM microcontrollers doing other tasks too, stuff like WiFi also includes an embedded cpu-core. Remember the X8 moniker for Moto X? That phone has a dual-core application processor, plus two other ARM cores, plus a couple they didn't count such as the baseband which can contain ARM cores (Intel's and several others does any way) now you have big.LITTLE with 8 cores too, plus the rest of the cores in the system. All from ARM basically. Older phones had like 2-3 cores, newer like 4-10 when it concerns ARM. Most of the sales are from the last five years most def. A few billions units quickly becomes those 37.5 billion, but you can pretty much conclude that those numbers includes wireless broadband cards and so on too.
If you just look at the Cortex-A graph, you see that they basically has come up to two billion a year by 2013, less than a billion a year for 12, a couple of hundred for 2011, an just tens of millions in 2010. Yet they didn't sell close to two billion devices with Cortex-A in 2013. Many system there was dual-core, quad-core and so on. Some were single-core, like chips for wifi routers or stuff for older devices.
Penti, you should stop being rude and read the first paragraph. Anand specifically emphasizes that the numbers represent discrete ICs containing ARM IP.
cumulative... I've owned an iPhone, iPhone 4, and iPhone 5. And an iPad and iPad3.
Additionally it's possible that every one of these phones (even if you count a dual CPU SOC as just one) had a separate low-power ARM running the RF part of the system. Once you get to iPhone 5S you explicitly have a low-power separate core handling sensors on a separate SOC, and other phones may have had something similar for a while, just didn't talk about it.
"A smart man once told me that no one wins by betting against performance."
For the desktop/server markets this is indeed very wise. In mobile, the metric to avoid betting against is power consumption and that's one of the key reasons why ARM is in 50 billions chips. Throughout their history, there has always been some one faster than ARM but not necessarily better performance per watt.
I am guessing the "A smart man once told me that no one wins by betting against performance." statement came from x86 world, AMD maybe?
Although I admit I now own more AMD cores than intel ones, albeit for the wrong reason. Ebay-ing old nehalems, yorkfields, ivy bridges recovers a lot more capital than selling old brazos... :(
The catch is performance per watt. If the amount of power being consumed is higher than what can comfortably run in a smart phone, it really doesn't matter how much faster a chip is at that point as it simply will not work in that form factor.
Isn't recycling to recover metals (gold contacts etc.) economically viable (in countries with cheap labor and weak occupational safety authorities)? Of course consumers probably mix broken phones with normal household waste but I'd imagine enterprise and embedded to pay a bit more attention to disposal, especially if they can sell the waste to recyclers.
It would be interesting to place this number in context compared to MIPS and PowerPC. As far as I can tell, MIPS and PPC do substantially better than ARM in highish end networking equipment, from cell tower equipment to network appliances. I THINK (with little to back it up) that they're also stronger in printers.
I don't know where or if Freescale give their shipment numbers, but It's interesting to see that their revenue is split pretty much evenly across micro controllers, networking, automotive and "analog & sensors", about $800M annually in each segment. I'm guessing that micro controllers equals printers and hard drives, and I'm surprised that auto is as large as it is --- maybe those are specialized CPUs hardened for heat and used to control the fuel injection?
While not a large market, I knew an engineer who was working with PowerPC equipment in avionics. I've been told that PowerPC is still popular in the automotive industry.
ARM IS doing this. That's the point of Linaro and SBSA.
You could wish that this had all happened sooner, but everything's running eighteen months later than it should because of ARM's bizarre delay in getting into the 64-bit game. However it is happening, and better late than never.
-- A smart man once told me that no one wins by betting against performance.
Well, that was true before Good Enough Computing became the law of the land. There's a reason so many PC/devices still sit on XP. Further, since there are vewy few embarrassingly parallel problems (still) and clocks are about as fast as one can get without melting the chip, it's going to be interesting to see whether there's much sequential performance left to be harvested and/or the compiler writers can turn purely sequential code into optimal parallel instructions.
I'm betting that we're near, or in, a vewy long pewiod of stasis.
Intel must not be very smart then, because they've been focusing mainly on power consumption to "catch up" with ARM. The irony is that by the time they do that, ARM will catch up with them in performance, if they launching Broadwell and Skylake with barely any *overall* performance improvements over Sandy Bridge, as they try to push power consumption lower and lower.
They lost in the same way with Atom, which used to be several times more powerful than the highest-end ARM chip, and now it barely competes on the CPU side, and that's with half a node advantage + Trigate, and it's a full generation behind in graphics performance (maybe more, compared to mobile Kepler).
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
34 Comments
Back to Article
jeffkibuule - Monday, March 31, 2014 - link
I'm guessing this doesn't include custom core SoCs like A6+ and Scorpion/Snapdragon.r3loaded - Monday, March 31, 2014 - link
It doesn't as it only counts shipments of ARM's processor IP rather than any architecture licenses.errorr - Monday, March 31, 2014 - link
I think your wrong. The Cortex-A are the custom chips that use the instruction set but aren't a specific design. Just like the Cortex-M are just IP blocks in things like modems and other chips throughout any device.extide - Monday, March 31, 2014 - link
No, Cortex A is a specific line, and things like Krait, and Apples Swift and Cyclone are NOT Cortex A, even though they are ~compareable. Stuff like the MTK, Allwinner, Exynos, and some of Quallcomm's designs that use the stock ARM IP (Like Snapdragon 610 and 400) are Cortex A. Also, some of the earlier Apple SOC's like the A4 and A5 (which were Cortex A8 and A9, respectively)blanarahul - Monday, March 31, 2014 - link
"ARM Partners" Ship 50 Billion Chips Since 1991 - Where Did They Go?ARM Partners. This includes Apple, Qualcomm everyone.
coder543 - Tuesday, April 1, 2014 - link
No... ARM doesn't make any chips. ARM just sells designs, so while it does include Qualcomm and Samsung, I would strongly lean towards the interpretation that these numbers only reflect ARM silicon shipments. Custom from-scratch designs like Apple's A7 or Qualcomm's Krait probably are not reflected in the data. Both Samsung and Qualcomm have shipped chips that use ARM designs, and both have shipped chips that do not use ARM designs but are ARM compatible. (Remember: Apple does not manufacture chips. Apple gets Samsung to manufacture their chips, traditionally.)FunBunny2 - Tuesday, April 1, 2014 - link
-- Custom from-scratch designs like Apple's A7Not really. As the other posting shows, Apple *assembles* the various blocks of ARM architecture into an A7. That's hardly "from scratch". An English major likely couldn't do it, but a competent EE sure could.
hypopraxia - Friday, April 4, 2014 - link
Ummm...you are dead wrong. The cyclone core you mention is ARMv8 instruction set compatible, but is by absolutely no means assembled from off the shelf IP blocks. Did Apple license a GPU for inclusion in the A7, yes. Did they license a CPU for inclusion in the A7, no.SeleniumGlow - Monday, March 31, 2014 - link
With this article I am now interested in seeing if Intel too follows the lisencing of IP route. Going by the way ARM is going, this is what appears to be the logical route, so so I thinkDribble - Monday, March 31, 2014 - link
Look at the turnover of ARM, look at the turn over of Intel. It's completely impossible for them to do what ARM does. Ironically they exist that way because they could not compete with Intel - instead of turning into AMD mk2 then stayed very lean and mean, living off $0.5 profit/chip. They went for the market Intel didn't want (embedded) and didn't attempt to control the market (like Intel and x86) but licensed designs to everyone.This is much more appealing to everyone using their chips as it's cheaper and gives them more freedom. Hence Intel's great problems getting into the ARM markets - even if they can produce competitive chips Intel can't survive on the tiny profits ARM makes, and no one wants to go back to being controlled by Intel anyway so will often license ARM in preference to buying x86.
grahaman27 - Monday, March 31, 2014 - link
CHARTS! FRAGMENTATION!I kid.
jjj - Monday, March 31, 2014 - link
ARM entering the server market with small cores has always been just the starting point, the intention to go bigger was always there.Ofc the industry is slow ,first they'll need to gain some market to develop the software ecosystem and might take some time before we see big cores. On the desktop side i keep saying that i wish the ARM guys would define some new standards and have a go at it.Wouldn't it be nice to start with a clean slate and modernize the desktop? Oh well, maybe AMD will sell us the server chips ,maybe Denver is good ,maybe Steam OS gets some ARM machines, or maybe not.StormyParis - Monday, March 31, 2014 - link
60% of 50 billion is 30 billion for mobile, that doesn't make sense any way you slice it, unless we're talking cores, not processors.grahaman27 - Monday, March 31, 2014 - link
since 1995? seems reasonable. How many times have you purchased an ARM-based device in the past 20 years? personally I've purchased at least a dozen, and I am just one person.Penti - Monday, March 31, 2014 - link
It makes perfect sense, a common smart phone uses around 10-15 ARM-cores today, and has always used at least 2-3 ARM-cores back to the feature phone days. ARM licenses cores. The processor IP is per unit (so that can be say 10 per chip) not per chip. IP is the keyword here.HammerStrike - Monday, March 31, 2014 - link
From the first paragraph of the article:"The bigger story was that the 10 billion in 2013 brought the cumulative total for ARM based processors to over 50 billion (note that these are discrete ICs, multiple cores within a single design are not counted multiple times)."
They are not counting individual cores, but discrete IC's. Granted, there may be a couple discrete IC's in each smartphone that ARM is licensing, but it is not an inflated count based on cores.
Latest data I found was 1 billion Android activations as of Sept 2013, with 1.5 million activations/day as of April 2013. So say in the last year they did an extra 750 million device activations (assumes activations/day continues to increase), and maybe another 500 million IOS devices that licensed the specific cores, and you have 2.25 billion "smart" phones. Say they have 3 IC's each on average licensed from ARM (no idea if that is a reasonable number or not, but that's what I'm going with), and you have a total of 6.75 billion chips from smart phones. If that math is within an order of magnitude that means ll the Nokia, Motorola, etc feature phones from the mid-90's on still make up the vast majority of their mobile total. But that would seem to be at odds from Anands statement that "37.5 of the 50 billion chips shipped in the past five years (2009 - 2013)". Either my estimates of smart phones/tablets totals is way off, the amount of licenses ARM gets per unit is significantly underestimated, or there is still a ton of feature phones shipped in the last 5 years.
No longer sure of the overall point of this post, but it's already typed, so I'll submit it.
Death666Angel - Monday, March 31, 2014 - link
Data I found (smartphone shippments):2010: ~300 m
2011: ~500 m
2012: ~650 m
2013: ~ 1 b
= 2.45b
Add to that some few hundred million tablet shipments, laptops, who knows how many dumb and feature phones....
Penti - Monday, March 31, 2014 - link
Read the damn graphs, it clearly says IP shipments! Each ARM core is an IP shipment. If you count IC's then a modern phone basically has a two, three or four ARM-powered chips. In reality these represents over 10 IP licenses from ARM. Just a modern application processor with built modem has at least something like 5-10 ARM cores. Nowadays there is usually a couple of ARM microcontrollers doing other tasks too, stuff like WiFi also includes an embedded cpu-core. Remember the X8 moniker for Moto X? That phone has a dual-core application processor, plus two other ARM cores, plus a couple they didn't count such as the baseband which can contain ARM cores (Intel's and several others does any way) now you have big.LITTLE with 8 cores too, plus the rest of the cores in the system. All from ARM basically. Older phones had like 2-3 cores, newer like 4-10 when it concerns ARM. Most of the sales are from the last five years most def. A few billions units quickly becomes those 37.5 billion, but you can pretty much conclude that those numbers includes wireless broadband cards and so on too.If you just look at the Cortex-A graph, you see that they basically has come up to two billion a year by 2013, less than a billion a year for 12, a couple of hundred for 2011, an just tens of millions in 2010. Yet they didn't sell close to two billion devices with Cortex-A in 2013. Many system there was dual-core, quad-core and so on. Some were single-core, like chips for wifi routers or stuff for older devices.
AndyKH - Wednesday, April 2, 2014 - link
Penti, you should stop being rude and read the first paragraph. Anand specifically emphasizes that the numbers represent discrete ICs containing ARM IP.name99 - Monday, March 31, 2014 - link
cumulative...I've owned an iPhone, iPhone 4, and iPhone 5. And an iPad and iPad3.
Additionally it's possible that every one of these phones (even if you count a dual CPU SOC as just one) had a separate low-power ARM running the RF part of the system. Once you get to iPhone 5S you explicitly have a low-power separate core handling sensors on a separate SOC, and other phones may have had something similar for a while, just didn't talk about it.
I think it's feasible.
Kevin G - Monday, March 31, 2014 - link
"A smart man once told me that no one wins by betting against performance."For the desktop/server markets this is indeed very wise. In mobile, the metric to avoid betting against is power consumption and that's one of the key reasons why ARM is in 50 billions chips. Throughout their history, there has always been some one faster than ARM but not necessarily better performance per watt.
PEJUman - Monday, March 31, 2014 - link
I am guessing the "A smart man once told me that no one wins by betting against performance." statement came from x86 world, AMD maybe?Although I admit I now own more AMD cores than intel ones, albeit for the wrong reason.
Ebay-ing old nehalems, yorkfields, ivy bridges recovers a lot more capital than selling old brazos... :(
dylan522p - Monday, March 31, 2014 - link
Silvermont atom wins in performance to watt vs anything ARM puts out. (only apple beats them but that's only slightly)Kevin G - Monday, March 31, 2014 - link
The catch is performance per watt. If the amount of power being consumed is higher than what can comfortably run in a smart phone, it really doesn't matter how much faster a chip is at that point as it simply will not work in that form factor.Wilco1 - Thursday, April 3, 2014 - link
If that were true, then everybody would be making phones with Silvermont in them. Where are they???landerf - Monday, March 31, 2014 - link
I was expecting something along the lines of how many are now in a landfill.Jeffrey Bosboom - Monday, March 31, 2014 - link
Isn't recycling to recover metals (gold contacts etc.) economically viable (in countries with cheap labor and weak occupational safety authorities)? Of course consumers probably mix broken phones with normal household waste but I'd imagine enterprise and embedded to pay a bit more attention to disposal, especially if they can sell the waste to recyclers.ruthan - Monday, March 31, 2014 - link
Pls add to those graph intel, i would be realy interesing.name99 - Monday, March 31, 2014 - link
It would be interesting to place this number in context compared to MIPS and PowerPC.As far as I can tell, MIPS and PPC do substantially better than ARM in highish end networking equipment, from cell tower equipment to network appliances. I THINK (with little to back it up) that they're also stronger in printers.
I don't know where or if Freescale give their shipment numbers, but It's interesting to see that their revenue is split pretty much evenly across micro controllers, networking, automotive and "analog & sensors", about $800M annually in each segment. I'm guessing that micro controllers equals printers and hard drives, and I'm surprised that auto is as large as it is --- maybe those are specialized CPUs hardened for heat and used to control the fuel injection?
Kevin G - Monday, March 31, 2014 - link
While not a large market, I knew an engineer who was working with PowerPC equipment in avionics. I've been told that PowerPC is still popular in the automotive industry.name99 - Monday, March 31, 2014 - link
ARM IS doing this. That's the point of Linaro and SBSA.You could wish that this had all happened sooner, but everything's running eighteen months later than it should because of ARM's bizarre delay in getting into the 64-bit game. However it is happening, and better late than never.
FunBunny2 - Monday, March 31, 2014 - link
-- A smart man once told me that no one wins by betting against performance.Well, that was true before Good Enough Computing became the law of the land. There's a reason so many PC/devices still sit on XP. Further, since there are vewy few embarrassingly parallel problems (still) and clocks are about as fast as one can get without melting the chip, it's going to be interesting to see whether there's much sequential performance left to be harvested and/or the compiler writers can turn purely sequential code into optimal parallel instructions.
I'm betting that we're near, or in, a vewy long pewiod of stasis.
MrRez - Wednesday, April 16, 2014 - link
Were did they go? Most of them would be in a landfill somewhere am guessing. wonder where the next 50 billion will end up?Krysto - Saturday, May 17, 2014 - link
Intel must not be very smart then, because they've been focusing mainly on power consumption to "catch up" with ARM. The irony is that by the time they do that, ARM will catch up with them in performance, if they launching Broadwell and Skylake with barely any *overall* performance improvements over Sandy Bridge, as they try to push power consumption lower and lower.They lost in the same way with Atom, which used to be several times more powerful than the highest-end ARM chip, and now it barely competes on the CPU side, and that's with half a node advantage + Trigate, and it's a full generation behind in graphics performance (maybe more, compared to mobile Kepler).