Apple A10 Fusion is ** Quad-core big.LITTLE **

Page 3 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.
Mar 10, 2006
11,715
2,012
126
Yep. Its far faster than i expected. And it looks like a very expensive solution.

What is all that cpu power good for?

Are they preparing to use it in laptops with this high freq stuff?? 4c a10 at 2.4 is sure damn fast.

To me there looks to be a huge difference in cost from say a 2 wide a73 plus tiny in order a53/a35 vs this monster. Quite different routes. Arm is clearly going the lean way while this looks more like core perf.

It's a premium phone. You don't cheap out on a premium phone.
 
  • Like
Reactions: NTMBK
Mar 10, 2006
11,715
2,012
126
Did you see that somewhere, or is that just a guess? It should be noted that A9 (which also includes embedded M9) is 105 mm2 on TSMC 16 nm.

That is a guess. I took NVIDIA GP104 die (314mm^2 on 16FF+) and adjusted for transistor count (7.2B versus 3.3B).
 

swilli89

Golden Member
Mar 23, 2010
1,558
1,181
136
It's a premium phone. You don't cheap out on a premium phone.

Except when you offer for years, 16gb as the minimum when everyone else has moved on. So I guess they really do cheap out on certain things, Apple is the worst.
 

Andrei.

Senior member
Jan 26, 2015
316
386
136
You can't change channel length with finFETs. FInFETs come in only 1 flavor AFAIK (whereas with planar x'tors designers can change certain aspects of x'tor feature sizes). That's one of the ways in which Intel improved density with 14nm. Higher performance per fin = less fins per transistors.

I'm not sure there's a change in transistor. Libraries are higher-level, right?

http://download.intel.com/newsroom/kits/22nm/pdfs/22nm-details_presentation.pdf
http://www.intel.com/content/dam/ww...ilicon-technology-leadership-presentation.pdf
Sorry I meant change path length, not channel. You can change transistors on processes like Samsung's where you have the option between the 78nm density and 84nm performance gate pitch libraries but most of it is also higher level (vendors can use their custom libs)

On the transistor level Intel also has the differentiation where it uses some high-speed transistors which are larger, I forgot their exact name right now but they had a pdf mentioning it when they were arguing their density non-advantage by comparing overall usage between the mobile SoCs and their CPUs. It was a presentation from earlier this year IIRC.

Edit: Short cell/tall cell: http://files.shareholder.com/downlo...A34CD/2015_InvestorMeeting_Bill_Holt_WEB2.pdf
 

witeken

Diamond Member
Dec 25, 2013
3,899
193
106
Sorry I meant change path length, not channel. You can change transistors on processes like Samsung's where you have the option between the 78nm and 84nm gate pitch but most of it is higher level.

On the transistor level Intel also has the differentiation where it uses some high-speed transistors which are larger, I forgot their exact name right now but they had a pdf mentioning it when they were arguing their density non-advantage by comparing overall usage between the mobile SoCs and their CPUs. It was a presentation from earlier this year IIRC.
You mean this one?

http://intelstudios.edgesuite.net/im/2015/pdf/2015_InvestorMeeting_Bill_Holt_WEB2.pdf

Apple's SoC looks to be higher density, but that's because they use devices with higher transistor density, like SRAM.
 

witeken

Diamond Member
Dec 25, 2013
3,899
193
106
My point was the short cell / tall cell difference mentioned there.
But couldn't that be because of the interconnect?

I remember, I think, when Idontcare was still here that he said once that there aren't as many x'tors on a die as you could place. (NAND has like 10x higher density, or so.) IIRC
 

Andrei.

Senior member
Jan 26, 2015
316
386
136
But couldn't that be because of the interconnect?

I remember, I think, when Idontcare was still here that he said once that there aren't as many x'tors on a die as you could place. (NAND has like 10x higher density, or so.) IIRC
From https://www.google.com/patents/US6385761 :
The cells in the library are defined by cell library definitions. Each cell library definition includes cell layout definitions and cell characteristics. The layout definition includes a layout pattern of the transistors in the cell, geometry data for the cell's transistors and cell routing data. The cell characteristics include a cell propagation delay and a model of the cell's function. The propagation delay is a function of the internal cell delay and the output loading of the cell.

Most cell libraries have cells that conform to a single cell height to allow abutment of adjacent cells into rows without creating design rule violations. The cell height is selected to achieve a desired balance between transistor performance and density for the integrated circuit. If the cell height is short, the transistors in the cells are smaller and have less drive strength. However, the rows of cells can be place closer together (i.e., at a smaller row pitch). This allows more transistors to be placed in a given area. Thus, a short cell height provides higher density, with lower performance. If the cell height is tall, the transistors in each cell are larger and have greater drive strength. However, the rows of cells must be placed farther apart such that fewer transistors can be placed in a given area. Thus, a tall cell height provides lower density, with higher performance. As a result, a single library of cells is often a compromise between performance and density.
Seems to be purely a library thing instead of a fixed process characteristic. Also matches up with what I've been told.
 

Phynaz

Lifer
Mar 13, 2006
10,140
819
126
Except when you offer for years, 16gb as the minimum when everyone else has moved on. So I guess they really do cheap out on certain things, Apple is the worst.

This is a really lame response. Apple didn't force anybody to buy the 16GB version, it was mearly an offering. As you said, it was the minimum.

I don't know how offering options to consumers makes a company "the worst".
 

Eug

Lifer
Mar 11, 2000
23,583
996
126
Who knows if this is legit or not, but the day after the launch event, I found this Geekbench 4 result, presumably for the iPhone 7 Plus:

https://browser.primatelabs.com/v4/cpu/213223

Operating System iOS 10.0.1
Model iPhone9,4
Processor ARM @ 2.23 GHz
2 processors
Processor ID ARM
L1 Instruction Cache 64 KB
L1 Data Cache 64 KB
L2 Cache 3072 KB
L3 Cache 0 KB
Motherboard D111AP
Memory 2998 MB

Single-core: 3233
Multi-core: 5363

For reference, the other iPhone 7 was here:

https://browser.primatelabs.com/v4/cpu/146939

iPhone9,3
2000 MB
3379/5495
 
  • Like
Reactions: Andrei.
Mar 10, 2006
11,715
2,012
126
Who knows if this is legit or not, but the day after the launch event, I found this Geekbench 4 result, presumably for the iPhone 7 Plus:

https://browser.primatelabs.com/v4/cpu/213223

Operating System iOS 10.0.1
Model iPhone9,4
Processor ARM @ 2.23 GHz
2 processors
Processor ID ARM
L1 Instruction Cache 64 KB
L1 Data Cache 64 KB
L2 Cache 3072 KB
L3 Cache 0 KB
Motherboard D111AP
Memory 2998 MB

Single-core: 3233
Multi-core: 5363

For reference, the other iPhone 7 was here:

https://browser.primatelabs.com/v4/cpu/146939

iPhone9,4
2000 MB
3379/5495

Absolutely insane amount of performance. Apple is well on its way to becoming the world's best client microprocessor design house...
 
  • Like
Reactions: NTMBK

Eug

Lifer
Mar 11, 2000
23,583
996
126
If it is legit, I'm actually happier more about the 3 GB RAM (for the Plus) than I am about the CPU performance.

Geekbench4_iPhone94_zpsufnz0bdy.png
 
Last edited:
Mar 10, 2006
11,715
2,012
126
If it is legit, I'm actually happier more about the 3 GB RAM (for the Plus) than I am about the CPU performance.

Geekbench4_iPhone94_zpsufnz0bdy.png

3GB of RAM for 7Plus has been rumored for a long time. Apparently needed for all of the fun stuff the dual lens camera does.

Looking forward to pre-ordering a 7 Plus tomorrow morning.
 

witeken

Diamond Member
Dec 25, 2013
3,899
193
106
Absolutely insane amount of performance. Apple is well on its way to becoming the world's best client microprocessor design house...
To validate that claim, one would need an independent in-depth investigation of Intel vs Apple, which goes beyond simple benchmarks and compares the technology at the level of for instance http://download.intel.com/newsroom/kits/14nm/pdfs/Intel_14nm_New_uArch.pdf, which we don't have. For instance, we know that Intel has this fancy Speed Shift technology, but what's Apple's (nameless) equivalent of that? How about FIVR? How about board area and thickness? How about DCC (Duty Cycle Control)? How about the LLC?

For benchmarking purposes, I would like to see the Apple architecture in a desktop configuration with frequency/voltage curves (like Idontcare has done), power consumption, overclocking, etc. Can A10 reach 5GHz :D?

In any case, you can't really say Apple makes the best microprocessors when their chips are a measly 4W TDP. But it helps for them that they're designing chips that the only go in >$600 products. And they have Anand Shimpi as head of their CPU group :eek:.
 

Eug

Lifer
Mar 11, 2000
23,583
996
126
3GB of RAM for 7Plus has been rumored for a long time. Apparently needed for all of the fun stuff the dual lens camera does
Yeah, but rumours are just that, rumours. It's always nice to have confirmation... although we still can't definitively say this is confirmation.

Looking forward to pre-ordering a 7 Plus tomorrow morning.
Yeah, that's the one I'm going to order, but what has given me pause is the reportedly crappy home button. A lot of people really don't like the new haptic non-button, as it seems its implementation isn't anywhere near as nice as it is for the MacBook Pro trackpads.

Apple giveth (A10 Fusion, 3 GB RAM, camera) and Apple taketh away (home button, headphone jack).

For benchmarking purposes, I would like to see the Apple architecture in a desktop configuration with frequency/voltage curves (like Idontcare has done), power consumption, overclocking, etc. Can A10 reach 5GHz :D?
Of course, that makes no sense either. Perhaps you can say that Intel makes better desktop chips than Apple... since Apple doesn't make any desktop chips.

In any case, you can't really say Apple makes the best microprocessors when their chips are a measly 4W TDP. But it helps for them that they're designing chips that the only go in >$600 products.
Indeed... and compete against Atom, which doesn't seem to be doing very well.

And they have Anand Shimpi as head of their CPU group :eek:.
Seriously though, I wonder what his position is there.
 

Eug

Lifer
Mar 11, 2000
23,583
996
126
The other thing of note is that Apple just updated the storage sizes of the iPad Air 2 and the iPad mini 4, without changing anything else. So now we have a whole bunch of big chips in iOS devices:

iPhone 7: ??? 140-150 mm2 (A10 Fusion)
iPad Pro: 147 mm2 (A9X)
iPad Air 2: 128 mm2 (A8X)

I guess it wouldn't make much sense to upgrade the iPad Air 2 to the A9X, since then it would eat into the iPad Pro sales even more than it already does, using an expensive chip with performance that isn't really necessary for most people. And it wouldn't make much sense to change it to the much smaller A9 (as a hypothetical iPad Air 3), since the multi-core performance between A8X and A9 is basically identical (at least according to Geekbench 4). Sure, the single-core performance of A9 is better, but I'm not sure it would make much sense to have two near identical products with chips with similar yet still different performance profiles. So, they're sticking with A8X and the iPad Air 2, virtually unchanged since 2014.

I wonder if this was intentional, but at the time I thought the A8X's third core might have been an anomaly, possibly related to technical issues that couldn't otherwise be overcome by other means in such a short time frame.