Apple A14 - 5 nm, 11.8 billion transistors

Page 7 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

scannall

Golden Member
Jan 1, 2012
1,944
1,638
136
Will Apple provide SVE in their ARM-based HEDT machines?
Reading ARM's website, their targetted use of SVE seems to be HPC and data-centres, and the internet out to the edge, but not beyond.
However, they do support running tools such as Forge on home computers, to develop SVE applications on the cloud, in the face of Covid-19.
What ARM targets for their own designs, and what Apple designs aren't the same thing. As far as what Apple will actually ship, we'll have to wait and see. It won't be much longer though.
 

Eug

Lifer
Mar 11, 2000
23,583
996
126
What ARM targets for their own designs, and what Apple designs aren't the same thing. As far as what Apple will actually ship, we'll have to wait and see. It won't be much longer though.
Yeah. It should be very soon.

It’s now October and the retail chains have begun receiving marketing materials for the iPad Air 4, so we’re probably talking just days now.

 

Richie Rich

Senior member
Jul 28, 2019
470
229
76
Yeah. It should be very soon.

It’s now October and the retail chains have begun receiving marketing materials for the iPad Air 4, so we’re probably talking just days now.

Lets see what the Geekbench5 score will be. I expect 1550 - 1600 pts @ 2.8 GHz in single thread which will outperform any today's desktops (Intel 10900K @ 5.3 GHz is around 1450 pts).
 
  • Like
Reactions: name99 and Eug

Eug

Lifer
Mar 11, 2000
23,583
996
126
Year after year after year they manage to impress.
The Apple A14:

ttps://twitter.com/HansDeVriesNL/status/1312251291704479744?s=20View attachment 30860
Yes, very impressive, and it looks like my guess was in the right ballpark.


Oh these are the same March ones? Hmm...

The number I have issue with is the 1658 single core. I was guessing 15xx so this seems a tad too high. The 4068 seems reasonable though.
 

Eug

Lifer
Mar 11, 2000
23,583
996
126

Is this Metal score possible? Seems way too high.
Is it too high though? Metal score of A14 is about the same as A12Z (within a few percent).

Maybe they felt they needed to up their GPU game for the Arm Macs.

BTW, if these CPU and GPU scores are truly legit, they definitely would be more than enough for a fanless 12” MacBook.
 
Last edited:

Richie Rich

Senior member
Jul 28, 2019
470
229
76
GB5 1583 pts is even more by 21 pts than my prediction where I used official data for 40% uplift from A12.
It looks like almost there is no need to change table data :D

But wait. What about the frequency?
Tweet says 2.99 GHz which too high............... this looks like fake.
L1 data cache only 64 kB? A13 has 128 kB..... looks like fake
L2 cache only 4 MB? A13 has 8 MB ................. this is really suspicious.

I think it's probably fake and somebody tries to gain attention.

Pos
Man
CPU
Core
Year
ISA
GB5 Score
GHz
PPC (score/GHz)
Relative to 9900K
Relative to Zen3
1​
Nuvia​
(Est.)​
Phoenix (Est.)​
2021​
ARMv9.0​
2001​
3.00​
667.00​
241.0%​
194.1%​
2​
Apple​
A15 (Est.)​
(Est.)​
2021​
ARMv9.0​
1925​
3.00​
641.70​
231.8%​
186.8%​
3​
Apple​
A14​
Firestorm​
2020​
ARMv8.6​
1562​
2.80​
558.00​
201.6%​
162.4%​
4​
Apple​
A13​
Lightning​
2019​
ARMv8.4​
1332​
2.65​
502.64​
181.6%​
146.3%​
5​
Apple​
A12​
Vortex​
2018​
ARMv8.3​
1116​
2.53​
441.11​
159.4%​
128.4%​
 

Eug

Lifer
Mar 11, 2000
23,583
996
126
GB5 1583 pts is even more by 21 pts than my prediction where I used official data for 40% uplift from A12.
It looks like almost there is no need to change table data :D

But wait. What about the frequency?
Tweet says 2.99 GHz which too high............... this looks like fake.
L1 data cache only 64 kB? A13 has 128 kB..... looks like fake
L2 cache only 4 MB? A13 has 8 MB ................. this is really suspicious.

I think it's probably fake and somebody tries to gain attention.

Pos
Man
CPU
Core
Year
ISA
GB5 Score
GHz
PPC (score/GHz)
Relative to 9900K
Relative to Zen3
1​
Nuvia​
(Est.)​
Phoenix (Est.)​
2021​
ARMv9.0​
2001​
3.00​
667.00​
241.0%​
194.1%​
2​
Apple​
A15 (Est.)​
(Est.)​
2021​
ARMv9.0​
1925​
3.00​
641.70​
231.8%​
186.8%​
3​
Apple​
A14​
Firestorm​
2020​
ARMv8.6​
1562​
2.80​
558.00​
201.6%​
162.4%​
4​
Apple​
A13​
Lightning​
2019​
ARMv8.4​
1332​
2.65​
502.64​
181.6%​
146.3%​
5​
Apple​
A12​
Vortex​
2018​
ARMv8.3​
1116​
2.53​
441.11​
159.4%​
128.4%​
What were you predicting for the clock speed? A lot of people were predicting 2.8 GHz to 3.0 GHz. Also note that this is an iPad, not an iPhone.

I am hopeful this is legit but we shall see. FWIW, it does follow Apple’s benchmark leak pattern for previous Ax releases. It usually shows up days before the official product release, to generate buzz. I’m not sure if it’s due to Apple releasing it themselves or reviewers doing it inadvertently, but my guess is both. Apple often first, and then NDA’d reviewers shortly afterwards.

EDIT:

This A13 bench report is showing 4 MB L2 for example, not 8 MB. And the L1 data cache for A13 is listed at only 48 KB.


5F722CF9-BD13-4AB4-8362-E5CF2CD2C995.png
 
Last edited:

Hitman928

Diamond Member
Apr 15, 2012
5,177
7,628
136
What were you predicting for the clock speed? A lot of people were predicting 2.8 GHz to 3.0 GHz. Also note that this is an iPad, not an iPhone.

I am hopeful this is legit but we shall see. FWIW, it does follow Apple’s benchmark leak pattern for previous Ax releases. It usually shows up days before the official product release, to generate buzz. I’m not sure if it’s due to Apple releasing it themselves or reviewers doing it inadvertently, but my guess is both. Apple often first, and then NDA’d reviewers shortly afterwards.

EDIT:

This A13 bench report is showing 4 MB L2 for example, not 8 MB. And the L1 data cache for A13 is listed at only 48 KB.


View attachment 30866

Here's a link to A13 versus the new result:


1601732487858.png

If legit it's a ~5% increase in single core points per clock in GB5.
 
  • Like
Reactions: Tlh97

Richie Rich

Senior member
Jul 28, 2019
470
229
76
What were you predicting for the clock speed? A lot of people were predicting 2.8 GHz to 3.0 GHz. Also note that this is an iPad, not an iPhone.

I am hopeful this is legit but we shall see. FWIW, it does follow Apple’s benchmark leak pattern for previous Ax releases. It usually shows up days before the official product release, to generate buzz. I’m not sure if it’s due to Apple releasing it themselves or reviewers doing it inadvertently, but my guess is both. Apple often first, and then NDA’d reviewers shortly afterwards.
I used 2.8 GHz for my IPC calculation ................... this means 11% IPC increase from A13.
But with 3 GHz and 1583 pts = 527 pts/GHz ........ this means 5% IPC increase only.

That would be the lowest IPC gain in Apple's history.The lowest IPC gain had A10 Hurricane core with only 8% uplift.

I really doubt that A14 will have that low IPC uplift. So for me it's fake or frequency is wrong. Score itself is OK. Also L1 cache is wrongly reported as A12 and A13 both have 128 kB data L1. I guess frequency is probably wrongly reported.

I expect 3.0 GHz in ARM MacBooks for different SoC with lower transistor densisty. But not in iPhone or iPad with same HD silicon. AFAIK A12X didn't have higher frequency than iPhone's A12.
 

Gideon

Golden Member
Nov 27, 2007
1,608
3,563
136
I used 2.8 GHz for my IPC calculation ................... this means 11% IPC increase from A13.
But with 3 GHz and 1583 pts = 527 pts/GHz ........ this means 5% IPC increase only.

That would be the lowest IPC gain in Apple's history.The lowest IPC gain had A10 Hurricane core with only 8% uplift.

I really doubt that A14 will have that low IPC uplift. So for me it's fake or frequency is wrong. Score itself is OK. Also L1 cache is wrongly reported as A12 and A13 both have 128 kB data L1. I guess frequency is probably wrongly reported.

I expect 3.0 GHz in ARM MacBooks for different SoC with lower transistor densisty. But not in iPhone or iPad with same HD silicon. AFAIK A12X didn't have higher frequency than iPhone's A12.
The actual clock speed is around 3.0. Geekbench measures it during ST test, and it has not been wrong for any other Apple product.

You can you can check from the json output by adding ".gb5" to the URL:
 

Attachments

  • Screenshot_20201003-172706.jpg
    Screenshot_20201003-172706.jpg
    69.3 KB · Views: 16
  • Like
Reactions: Tlh97

Richie Rich

Senior member
Jul 28, 2019
470
229
76
The actual clock speed is around 3.0. Geekbench measures it during ST test, and it has not been wrong for any other Apple product.

You can you can check from the json output by adding ".gb5" to the URL:
OK, that means the lowest IPC uplift in Apple's history is real (5%).

On the other hand it's positive to have similarity with A10. Because A10 was the last 4xALU core and next A11 was completely new uarch with 6xALU (1st on the world). It looks like Apple has pattern of ground up uarch redesign every 4 years. This means next year Apple A15 will be brutal beast similar to Nuvia Phoenix (same chief architect G. Williams though), probably first 8xALU monster core on the world (funny when x86 are stuck at 4xALU design which Apple abandoned back in 2016).

Also ARMv9 ISA with SVE2 vectors might went into play for A15. Apple probably concentrated resources to work on brand new A15 for next year. Also a lot of work on interconnect for A15 with more than 16-cores as Xeon replacement. A lot of work for next year. So no wonder A14 is little polished and shrinked A13 only.

Funny is that A15 is probably taped out already and maybe even booting OS on early ES chips.
Funny is that A16 uarch will be finished soon and A17 and A18 are in design and thinking about concept of A19 new line up with 12xALU :D
 
  • Haha
Reactions: NTMBK

name99

Senior member
Sep 11, 2010
404
303
136
GB5 1583 pts is even more by 21 pts than my prediction where I used official data for 40% uplift from A12.
It looks like almost there is no need to change table data :D

But wait. What about the frequency?
Tweet says 2.99 GHz which too high............... this looks like fake.
L1 data cache only 64 kB? A13 has 128 kB..... looks like fake
L2 cache only 4 MB? A13 has 8 MB ................. this is really suspicious.

I think it's probably fake and somebody tries to gain attention.

Pos
Man
CPU
Core
Year
ISA
GB5 Score
GHz
PPC (score/GHz)
Relative to 9900K
Relative to Zen3
1​
Nuvia​
(Est.)​
Phoenix (Est.)​
2021​
ARMv9.0​
2001​
3.00​
667.00​
241.0%​
194.1%​
2​
Apple​
A15 (Est.)​
(Est.)​
2021​
ARMv9.0​
1925​
3.00​
641.70​
231.8%​
186.8%​
3​
Apple​
A14​
Firestorm​
2020​
ARMv8.6​
1562​
2.80​
558.00​
201.6%​
162.4%​
4​
Apple​
A13​
Lightning​
2019​
ARMv8.4​
1332​
2.65​
502.64​
181.6%​
146.3%​
5​
Apple​
A12​
Vortex​
2018​
ARMv8.3​
1116​
2.53​
441.11​
159.4%​
128.4%​

Historically GB5 has been OK at getting the frequency and lousy at getting the cache sizes. Look at their values for A13s. The frequency is reasonable for 5nm.

More analysis here:
 
  • Like
Reactions: Richie Rich

Eug

Lifer
Mar 11, 2000
23,583
996
126
Historically GB5 has been OK at getting the frequency and lousy at getting the cache sizes. Look at their values for A13s. The frequency is reasonable for 5nm.

More analysis here:
A4?

And while I don't claim to be very knowledgable about this stuff, I personally don't consider these numbers disappointing, just that we shouldn't expect a 20% IPC increase every generation anymore.

Plus, the GPU seems to have gotten a major boost this generation. This bodes well not only for Arm Macs, but also an Apple TV as game console.
 

Roland00Address

Platinum Member
Dec 17, 2008
2,196
260
126
BTW, if these CPU and GPU scores are truly legit, they definitely would be more than enough for a fanless 12” MacBook.
You are too easily satisfied Eug 😂 I personally demand more cores!

2 big + 4 little is not enough for a premium device that will cost $1k + with the right ram and ssd option. Let apple settle this dissatisfaction by offering
  • fanless 12" "cheap" with the A14 (2 big + 4 little)
  • fanless 12 "heavy" with the A14X (4 big + 4 little and better gpu)
...and charge another $200 more for the heavy option. Sure in single thread it will make no difference but cater to my ego or something 😉

-----

Apple has the tech for higher end skus then offer the higher price options so Apple can recap the most value from Apple customers who want to pay for it. Have a 2020 iPad Pro with the A14X and a 12" fanless macbook pro and a 12" fanless macbook cheap. Offer the skus (and good values) at every price brackets if you can segment the market with top end silicon. [Likewise the future 15" / 16" should have 8 big and 8 small cores and not be fanless allowing the chips to run up to 45w even if the real goal is something like 15w or 28w.]

(Said a person who is using a 2019 ipad 7th gen with an A10 for $249 last year and I am satisfied with its performance. (in Geekbench 747
Single-Core Score, 1375 Multi-Core Score, so we are talking 2x for Single Core, and 3x for Multi Core for the $599, that is a great value for the air even though the 2020 8th gen is now a12 with 1100 ish single core and 2800 ish multi core based on the last year's ipad air.)
 

Eug

Lifer
Mar 11, 2000
23,583
996
126
You are too easily satisfied Eug 😂 I personally demand more cores!

2 big + 4 little is not enough for a premium device that will cost $1k + with the right ram and ssd option. Let apple settle this dissatisfaction by offering
  • fanless 12" "cheap" with the A14 (2 big + 4 little)
  • fanless 12 "heavy" with the A14X (4 big + 4 little and better gpu)
...and charge another $200 more for the heavy option. Sure in single thread it will make no difference but cater to my ego or something 😉
Yeah, my needs in a laptop are pretty limited. My main concern at this point are:

1) Fanless
2) Dual USB-C ports.
3) Better keyboard
4) Better trackpad
5) Better FaceTime camera would also be nice.

However, I won't be upgrading my Core m3-7Y32 12" MacBook any time soon. It's not super fast, but it's fast enough that I have no complaints for speed, for my on-the-road business usage. My main complaint about it is the single USB-C port.

I do agree though it would be good to have two tiers of the fanless MacBook though. Not necessarily 4+4 though. Would a 3+4 be a feasible option?

However, I wouldn't at all be surprised if a fanless MacBook was 4+4, for marketing reasons. It'd more easily allow them to keep pricing high.


(Said a person who is using a 2019 ipad 7th gen with an A10 for $249 last year and I am satisfied with its performance. (in Geekbench 747
Single-Core Score, 1375 Multi-Core Score, so we are talking 2x for Single Core, and 3x for Multi Core for the $599, that is a great value for the air even though the 2020 8th gen is now a12 with 1100 ish single core and 2800 ish multi core based on the last year's ipad air.)
I bought an A10 iPad 7 cellular for the wife for cheap last year. Performance is fine on that too, for surfing, email, business applications, and its 3 GB RAM is relatively OK.

I am currently using an A10 iPhone 7 Plus with 3 GB RAM from 2016 and performance on that is fine too. I never notice any significant slowdowns related to CPU speed. I only have noticed occasional pauses which I attribute to RAM.

I'm looking to upgrade to an A14 iPhone 12 Pro or Pro Max this year mainly because I want the new camera, and it also will be nice to get 6 GB RAM. 5G will also be a bonus. Otherwise I'd be fine to keep this phone for another year.
 

Doug S

Platinum Member
Feb 8, 2020
2,201
3,405
136
You are too easily satisfied Eug 😂 I personally demand more cores!

2 big + 4 little is not enough for a premium device that will cost $1k + with the right ram and ssd option. Let apple settle this dissatisfaction by offering
  • fanless 12" "cheap" with the A14 (2 big + 4 little)
  • fanless 12 "heavy" with the A14X (4 big + 4 little and better gpu

I wouldn't necessarily assume the A14 will show up in any Mac, the iPhone is Apple's big money product and they wouldn't want to risk its schedule by including extra bits in the design only necessary for the Mac line.

Probably makes more sense to use the A14X (or whatever suffix letter(s) we may see) for every Macbook Air, and adjust the number of enabled cores / clock rate to create 2 or 3 levels of performance. Since they don't have a defined release schedule for iPad Pro like they do for the iPhone they won't care so much if its release slips by a few months because of some Mac related issues with the SoC.

I wouldn't be surprised if the top end CPU core/clock config serves as the low end for the next step up. That is high end Macbook Air = low end Macbook; high end Macbook = low end Macbook Pro.
 

Roland00Address

Platinum Member
Dec 17, 2008
2,196
260
126
I wouldn't necessarily assume the A14 will show up in any Mac, the iPhone is Apple's big money product and they wouldn't want to risk its schedule by including extra bits in the design only necessary for the Mac line.

Probably makes more sense to use the A14X (or whatever suffix letter(s) we may see) for every Macbook Air, and adjust the number of enabled cores / clock rate to create 2 or 3 levels of performance. Since they don't have a defined release schedule for iPad Pro like they do for the iPhone they won't care so much if its release slips by a few months because of some Mac related issues with the SoC.

I wouldn't be surprised if the top end CPU core/clock config serves as the low end for the next step up. That is high end Macbook Air = low end Macbook; high end Macbook = low end Macbook Pro.
*Nods*

That makes sense and I would be fine with die harvested A14X stuff or whatever letter they use for they may use different letters for the mac line. Like do you need pci-express on a macbook line? You do not need a gpu, wifi can use a different connection, etc? Do we get Thunderbolt 3 / USB 4? How about Thunderbolt 4? etc, etc.

I am also fine with apple put lots of those mini cores on there since they are so die efficient. I just want 4 big cores on any $1k price mac even if 2 big cores still perform very well. I am also fine with 2 big cores being a cheaper option for $800 to $1000 prices, even $1050. ( Yes it not about me, but we are talking the subjective experience of "satisfaction" here. )
 

Doug S

Platinum Member
Feb 8, 2020
2,201
3,405
136
*Nods*

That makes sense and I would be fine with die harvested A14X stuff or whatever letter they use for they may use different letters for the mac line. Like do you need pci-express on a macbook line? You do not need a gpu, wifi can use a different connection, etc? Do we get Thunderbolt 3 / USB 4? How about Thunderbolt 4? etc, etc.

I am also fine with apple put lots of those mini cores on there since they are so die efficient. I just want 4 big cores on any $1k price mac even if 2 big cores still perform very well. I am also fine with 2 big cores being a cheaper option for $800 to $1000 prices, even $1050. ( Yes it not about me, but we are talking the subjective experience of "satisfaction" here. )

It will be interesting to see what I/O they have. No chance Macbooks will have PCIe - I wouldn't even be 100% sure it would be included on the Mac Pro. Since Apple will be using their own GPU, and you will have multiple TB/USB ports if you need something like an external storage array or additional 10GbE interface, what would be the point of PCIe on the Mac Pro when almost no one would use it... I know, I know, cut the horrified gasps from people who say it can't be "Pro" without the slots, but tell me what you need them for if you can't add a GPU and a TB4 port wouldn't be fast enough.
 

Richie Rich

Senior member
Jul 28, 2019
470
229
76
Historically GB5 has been OK at getting the frequency and lousy at getting the cache sizes. Look at their values for A13s. The frequency is reasonable for 5nm.

More analysis here:
Nice analysis at RWT and I agree with #2 scenario: A14 is small IPC/uarch jump.
According to my IPC table the evolution of 4xALU Apple line up is like this:
  1. A7 Cyclone .... first 4xALU ARM core on the world, 2013, same year as 4xALU Intel Haswell, IPC similar to Sandy Bridge
  2. A8 Typhoon .... +11% over A7
  3. A9 Twister ...... +32% over A8 .... massive IPC jump (we know they doubled shifters 2x -> 4x, so my guess is Apple went from 4xALU with shared BranchU uarch to 4xALU+2xBr)
  4. A10 Hurrican ... +8% over A9 .... the least IPC jump (last 4xALU core) but big frequency jump (1.8 -> 2.3 GHz)

It's clear that:
  • Apple use 4-year cycle when they do new groud-up uarch.
  • One major uarch change after 2-years (A9 was massive IPC uplift)
  • last step uarch is weak due to uarch IPC potential was depleted, big frequency jump as compensation

When we take a look at their 6xALU uarch we can see similar pattern:
  1. A11 Monsoon ... +19% IPC over A10 (+41% over 9900K)..... 6xshifters so probably 6xALU + 2x Br (8-wide, double wide than AMD and Intel)
  2. A12 Vortex ... +13% IPC over A11 .... cache subsystem redesign, lower latencies, L1 64kB -> 128kB
  3. A13 Lightning ... +14 IPC over A12 ... added 3rd Branch pipe, so 6xALU+3xBr = 9-wide integer core (widening core like A9 did)
  4. A14 Firestorm ... +5% only (last 6xALU core) .... only +91% IPC over 9900K

No double IPC with A14 this year, only 1.91x :(


So A14 Firestorm is similar to A10 Hurricane:
  • low IPC jump due to all low hanging fruits form this uarch was eaten
  • big frequency jump as low IPC jump compensation
  • last core from this uarch line up
  • 12-core MacBook version might keep engineers busy with uncore stuff like interconnect and LLC
  • A15 as new 8xALU+3xBr uarch based on ARMv9 + SVE2 might consume a lot engineering resources as well
  • 32/64-core version (Xeon replacement in two years) might consume also some stuff (probably A16)
  • I don't think G.Williams III leaving in 2018 will hurt Apple's uarch dev. In 2018 GW3 were working on A15 and A16 already, those are his children still (I guess Nuvia Phoenix will be very similar to A16). So we might see some IPC slow down with A17 as this should be another wider uplift of A15 line up. Well'll see, they are working on A17 one year already and starting work on A18.
  • Mike Filipo will lead development of new A19 line up (A19,20,21,22). And he is even more radical in widening core. When you look at his Cortex A72, he separated not only ALU and Br but also multicycle DIV/MUL unit. His Cortex A78/X1/V1 is 1st gen developed after SoftBank purchase of ARM and it's wide beast. I'm not afraid about Apple's CPU future at all. I'm afraid about AMD and Intel, they can lose server/NTB/desktop market in less then 5 years and disappear like DEC Alpha, Cyrix, IDT/VIA, PowerPC, Itanium.
 
  • Like
Reactions: name99

Eug

Lifer
Mar 11, 2000
23,583
996
126
Is it too high though? Metal score of A14 is about the same as A12Z (within a few percent).

Maybe they felt they needed to up their GPU game for the Arm Macs.
It should be noted though that A14 is just 4-core for the GPU, like A13 and A12, which means it’s got only half the number of GPU cores of A12Z. Yet, the compute scores of A14 and A12Z are about the same.
 
  • Like
Reactions: Richie Rich