Apple A7 is now 64-bit

[AnitaPeterson]

Thread reminds me of 2003, A64 all over again.

"Who needs more than 4GB of ram pfft gtfo!"

Better to have and not need than need and not have. :thumbsup:

I feel old

You're forgetting a little detail: Mobile products are not upgradeable by the end-user. Real computers, on the other hand...

 

[Schmide]

Are you from a GBA homebrew community or something? XIP doesn't work with NAND flash, which is what most devices use for storage.

There is no reason it wouldn't work with NAND??? It's just a matter of mapping the file, via page table, into the appropriate space. The type of memory is always abstracted from the OS.

 

[blackened23]

Or they're on their way to ditching Intel entirely. Apple has an ARM architectural license so they could conceivably design their own high performance ARM chip and put them in MacBooks.

Yeah, except, no. Apple is not the same stratosphere as intel is w.r.t. performance, comparing A7 or any ARM SOC to core I5/I7 (for macbooks) is like comparing a mentally challenged, crippled amputee to an olympic athlete. Give me a break.

Yeah, Apple has a nice chip for phones. If you think you'll see this in macbooks anytime soon, that's a fantasy.

 

[Nothingness]

So it's a pretty safe bet this is TSMC 28nm right? I don't think Apple has ever reported transistor counts before, it sounds like they want to show it off. A good indication that the density is substantially better than it was with Samsung, which is something I've always strongly suspected.

I simply don't know much about process density so I won't hazard a guess

For a comparison, XBox One's SoC is 363mm^2 and about 5 billion transistors, making it even denser than A7 (it's also on TSMC 28nm). But a lot of it is SRAM.

Ivy Bridge also has a lot of SRAM on-chip (8 MB L3 cache, assuming Intel quoted transistors for 3770K).

 

[Rakehellion]

I think iPhone 5S is going to be quad core or possible triple core, because they claim up to 2x CPU performance over iPhone 5, and it's hard to believe they achieve that with the same core count.

They've doubled the speed every year and haven't gone over two cores yet, I don't see why.

 

[ShintaiDK]

I'm surprised no one noticed this: >1B transistors for 102mm². That's denser than 22nm Ivy Bridge with it 1.4B transistors and 160mm².

And less dense than a 28nm GK107 for example. Density doesnt say much.

 

[podspi]

http://www.samsung.com/global/business/semiconductor/news-events/press-releases/detail?newsId=12979

Yes, special memory just for the 32bit ARM limit.

Though I do agree that we'll probably see a temporary ceiling at 3gb due to ARM's 32-bit limitation, this hardly qualifies as developing 'special memory' just for the 32bit ARM limit. They also produce 1gb and 2gb chips. Nothing in that link supports your claim.

I'm surprised no one noticed this: >1B transistors for 102mm². That's denser than 22nm Ivy Bridge with it 1.4B transistors and 160mm².

Don't GPUs and SRAM tend to be denser than CPU logic? I would bet it is because the A7 has a huge GPU.

 

[Vesku]

ARM SoC performance and available RAM is creeping past the desktop P4 level so it seems logical to move to 64 bit. Kind of silly to rehash the whole X amount of RAM should be plenty for anyone. As was seen with the x86 64 bit transition it can lead to a better overall OS experience.

I won't turn my nose up at phones and tablets with 4GB+ of RAM combined with processors closing in on or even beating notebook CPUs from 2009-10.

 

[Exophase]

There is no reason it wouldn't work with NAND??? It's just a matter of mapping the file, via page table, into the appropriate space. The type of memory is always abstracted from the OS.

XIP needs memory that's actually part of your address space. You can't put NAND on your address space because you need to prepare a read for a page in advance of reading it. You could make an interface that does this implicitly but the performance would be awful. Much better to let the OS manage loading it into RAM via its virtual memory subsystem.

They've doubled the speed every year and haven't gone over two cores yet, I don't see why.

One of those times came from doubling the core count. Most of the other times were accompanied with process node shifts. And it's almost certain that they're not going to be looking at a major improvement in electrical characteristics of their process, at best they're going from Samsung 32nm to TSMC 28nm. Improving single threaded performance costs more power consumption than improving peak performance via more cores. You can't just pull a doubling of performance out of thin air on a not-that-much-better processor without costing a lot more power, and Apple has always been pretty miserly with CPU power consumption.

I don't think they're going to all of a sudden grow tremendously in perf/MHz over Swift, which already has good perf/MHz for its class. And I don't think they'll go up in clock a lot. But I guess we'll find out (I hope)

 

[Sheep221]

I thought that only x86 systems use 32 and 64-bit addressing(hence the x86-32 and x86-64 names).

 

[Nothingness]

I thought that only x86 systems use 32 and 64-bit addressing(hence the x86-32 and x86-64 names).

MIPS did that before x86.

 

[Arachnotronic]

I'm surprised no one noticed this: >1B transistors for 102mm². That's denser than 22nm Ivy Bridge with it 1.4B transistors and 160mm².

A7 is likely to turn out to be mostly GPU. GPUs are much "denser" than CPUs

 

[Exophase]

MIPS did that before x86.

SPARC as well. And POWER for that matter. And probably many others I don't know.

 

[Haserath]

A7 is supposedly dual core. Average 31% improvement.
http://www.geek.com/apple/iphone-5s...hone-5-includes-motion-tracking-chip-1568438/

The new swift core can also run at 1.7Ghz, but most likely will be lower for power reasons.
---
It would be nice to see a dockable phone with iOS-like and OSX-like operating systems in the future.

 

[VengenceIsMine]

SPARC as well. And POWER for that matter. And probably many others I don't know.

Alpha

 

[Nothingness]

Alpha

Alpha never was 32-bit

 

[StinkyPinky]

A7 is supposedly dual core. Average 31% improvement.
http://www.geek.com/apple/iphone-5s...hone-5-includes-motion-tracking-chip-1568438/

The new swift core can also run at 1.7Ghz, but most likely will be lower for power reasons.
---
It would be nice to see a dockable phone with iOS-like and OSX-like operating systems in the future.

I'm surprised someone hasn't already done this to be honest.

Maybe the limitation is the display output. Cramming that into a phone could be cumbersome.

 

[seitur]

I'm surprised someone hasn't already done this to be honest.

Maybe the limitation is the display output. Cramming that into a phone could be cumbersome.

Wut? Of course someone already did. You have docks for smartphones on market already. Just not for Apple. Hell you have ARM dongles connectable to big screens that are more than twice smaller than a smartphone. Soon you'll have such small x86 devices as well. There is no display output problem at all.

 

[erunion]

I'm surprised no one noticed this: >1B transistors for 102mm². That's denser than 22nm Ivy Bridge with it 1.4B transistors and 160mm².

probably because a higher portion of it is GPU.

 

[Vesku]

If Apple wasn't so brand focused they could offer 28nm 64 bit ARM HPC servers and beat the rest of the market by ~6-9 months.

 

[lopri]

64 bit ARM leaves me with conflicting feelings, it's quite possible that this will be the isa that the majority of consumers will use in the future.

On one hand more companies battling in the market is a positive thing compared to the x86 duopoly, but I can't see us consumers ever getting the same freedom we enjoy regarding OS upgrades in the PC market.

People are spending an increasing amount of money into devices with strong planned obsolescence "features" (locked bootloader, locked os, locked app platform) and this is not good.

This whole mobile race looks similar to the 80s pc market in many ways.

I think you are speaking to a wrong audience. Last time I checked Intel supporters on this board love Intel's 'smart' business tactics such as planned obsolescence, artificially crippled feature sets, forced upgrades via incompatibility, etc. They should feel right at home with recent developments.

 

[Rakehellion]

If Apple wasn't so brand focused they could offer 28nm 64 bit ARM HPC servers and beat the rest of the market by ~6-9 months.

How would that make them any money?

 

[Intervenator]

The reason why the A7 is 64-bit is to prepare for when a 64-bit SoC will actually be needed by Apple. By the time the iPhones and iPads of the future are ready for 4GB+ of RAM, most major apps on the App Store will be optimized for a 64-bit SoC. This will give Apple a major advantage over the competition because Apple will have substantially more apps that are a "generation" ahead of the competition. This is assuming that "the competition" does not develop 64-bit SoCs before phones with 4GB+ of RAM becomes relevant.

 

[StrangerGuy]

I'm much interested in the GPU. it's almost certainly going to be some flavor of PowerVR Rogue.

 

[2is]

Or they're on their way to ditching Intel entirely. Apple has an ARM architectural license so they could conceivably design their own high performance ARM chip and put them in MacBooks.

Not going to happen. They are nowhere near as powerful and would be nowhere near as efficient if they tried to make them anywhere near as powerful. Not to mention the huge customer base they would lose by not only not being backward compatible with their own software, but incompatible with Windows which a whole lot of mac users also run on their machines. (myself included)