Apple TV's A5 Chip Includes Surprise Die Shrink to 32nm [Updated: Same Chip Found in iPad 2]

[MythicFrost]

Actually they are cheaper because more die fit on each wafer, the die for the new A5, is 41% smaller then the die for the old A5. Smaller die, more die's per wafer and thus a cheaper part. The same A5s are being used in the Apple TV3, and the newer cheaper Ipad 2's, which means the second core is disabled in the firmware on the Apple TV 3.

Ah right, I see. So the wafer is a certain size, and they stuff as many A5's on them as possible, and it's cheaper because they can now fit more on, while still paying the same price for the wafer?

Silicon chips have random defects - it's the nature of the materials and the process.

A wafer has many chips on it - some will be defective, some will be OK.

If a dual-core chip has a defect in the chip area for one core, it may still be able to function perfectly well as a single-core chip.

In the case of a shrink like this, it's quite possible that the number of chips which work correctly as dual-core is fairly low - but that a good number of them work fine as single-core chips.
_________________

As an example, Intel's "Celeron" CPUs early on were just Pentiums with half-sized cache. They were actually from the same wafers as the Pentiums, but had defects in the cache. Intel's process could disable the defective regions of the cache, and sell the chips as Celerons with half the cache as a Pentium.

Ah right, I see.

 

[mrklaw]

why is it a surprise? Logically Apple will want to move to a smaller die size ASAP. Maybe they were trying for the 4S but couldn't get the yields they needed, so they used binned parts for the apple TV. Either that was the plan all along while they refine their processes to prepare for iphone 5, or its a smart reaction to make the most of available chips rather than dump them all

----------

Yes, they did. Every successive iphone from 3GS to 4GS had a speed bump of some sort. The 3GS was a 600 MHz Cortex A8 on a 65nm process. The A4 was a 1000 MHz (800 MHz) Cortex A8 on a 45nm process. The A5 was a 1000 MHz (800 MHz) dual core Cortex A9 on a 45nm process. The iPhone to iPhone 3G transition was the only time the iPhone didn't get a CPU bump.

And they are using the fully functional version on the iPad 2 now, and will likely use it for the 4S as well.

It actually makes more sense to make a 32nm Cortex A15 chip for the next iphone, as that CPU will be more efficient at a given clock speed, getting the task done faster and being able to idle more.

do you think they'll roll this into 4S's during the current active lifecycle, or switch to it when the iphone 5 comes out (assuming the 4S becomes the lower priced entry, like they are doing with the 'new' ipad 2)

I agree I don't think they'd use this for iphone 5. If the process is stable enough they could maybe run an A5X through it for power saving, or even something else (A15 is a nice wish to have)

 

[Colpeas]

If u dont care how its achieved buy an ipod classic. 160gb. Sold at the apple store as we write.

Maybe Off-Topic, but yeah, we also need new iPod Classic. 160GB is becoming insufficient for my needs - I am a huge lossless audio fan. and I still don't give a damn how Apple will cope with stuffing such storage inside. The whole problem is that nothing is pushing them to update it. Classic *needs* some competition.

 

[whooleytoo]

What purpose would there be even if it was possible? the Apple TV doesn't do anything remotely close to needing dual core. What do you want to do, play 4 movies at the same time?

Maybe more CPU power would be nice for greater video compression? For me scrubbing through HD videos on my ATV is a jumpy, buggy process. Lower bandwidth requirements and a faster CPU to process the video might help.

Of course, you'd either have to re-encode your videos, or politely ask Apple/Youtube/Netflix etc. to re-encode their video libraries...

 

[justperry]

THANK YOU SAMSUNG!!!
I truly thank you on behalf of all blind followers. Thank you for your new 32nm Samsung High-K Metal Gate manufacturing process. You are leaders in innovation and invention. Without you, none of this would be possible. Truly magical.

Thank you ASML.
Thank you ARM.
Thank you Einstein.
Thank you for the sun..
Thank you BIG Bang.
Thank you ....................................................."God" (I-Atheist)

As if Samsung is the only one that innovate, invent manufacture ASO.

 

[chrmjenkins]

Remember, Apple is still making the iPad 2 ($400), that means they plan to support the A5 for awhile.

Well, A5 support is a given as long as the apple TV exists. That product can exist unchanged for years, so there's no question A5 is here to stay.

Also keep in mind the New iPad is exactly the same as the iPad 2 except for more RAM and quad graphics to support the Retina screen. I would say the iPhone 5 will be a A5, and I wonder if Apple will care to put more RAM in it, since the extra RAM in the New iPad is for graphics (more RAM means more power drain). However if Apple does add more RAM to the iPhone 5 then I think it will just be 768MB.

It all depends how ambitious iOS 6 is and how aggressive they want to be with their technical features. Historically, they've been very aggressive with their SoCs, and I don't see any reason for that trend to not continue. It's pretty clear the A5X isn't a good fit for the next iPhone, so they'd have to cook up something new.

The iPad 2, New iPad, iPhone 4S, and iPhone 5 will support iOS6. I give a %75 chance that the iPhone 4 (they are still selling the iPhone 4 too) and original iPad will support iOS6, but performance will definitely take a hit.

I would say there is a 100% chance they support iOS 6, minus some features. The iPhone 4 will only be 2 generations old at that point. They've always supported 2 generation old hardware in a new iOS release. iPhone got iOS 3, iphone 3G got iOS 4, 3GS got iOS 5, etc.

The iPhone 5 will sell well, because of redesigned look (bigger screen), and LTE. The iPhone 5S will sell well because of A6, etc. iPad 4 will be first iOS device with A6.

People seem to think this bigger screen is a foregone conclusion. The form factor and 3.5" has been a constant of one of the most successful mobile products ever. I don't see a big reason for them to change it. The next iPhone can get a dual core cortex A15 because there's still places to go after that. Quad-core, big.LITTLE, at some point they step up to LPDDR3, Rogue graphics cores, die shrinks etc. etc.

do you think they'll roll this into 4S's during the current active lifecycle, or switch to it when the iphone 5 comes out (assuming the 4S becomes the lower priced entry, like they are doing with the 'new' ipad 2)

I agree I don't think they'd use this for iphone 5. If the process is stable enough they could maybe run an A5X through it for power saving, or even something else (A15 is a nice wish to have)

Sure, they'll do it as soon as capacity and yield allow and it makes financial sense. They may be on a supply curve where pushing the supply into the 4S also would actually make it cost more per chip. I'd say it will be in there by the time we get the next iphone though. I also don't see a need for A5X in the next iPhone. It's simply too much of a brute force approach for more GPU power. A rogue dual core would much more elegantly achieve the same performance (actually more).

 

[SBlue1]

More battery time and less heat? Would be great.

 

[Rocketman]

It gets bad after 28nm.

Interesting chart. This indicates to me the current chip release may have particularly long legs, perhaps 3 years or more. 2012-13, 2013-14, 2014-15 and probably into 2015-16 too. Getting people used to that idea would reduce the whining.

Rocketman

 

[chrmjenkins]

Interesting chart. This indicates to me the current chip release may have particularly long legs, perhaps 3 years or more. 2012-13, 2013-14, 2014-15 and probably into 2015-16 too. Getting people used to that idea would reduce the whining.

Rocketman

People are starting to say the limit for silicon is somewhere in the 7nm range. Intel's FinFET and these high-k gate first technologies are only the beginning of the tricks we'll need to get there. Roadmaps don't go past 4nm. At that point, you're talking a matter of single-digit atoms wide anyway. We're within a decade of the end of that roadmap http://en.wikipedia.org/wiki/Semiconductor_device_fabrication

As far as I can tell, graphene will be the next great enabler, but it's still quite a ways off. Too tough to make right now.

Short Q&A on graphene's future: http://nextbigfuture.com/2011/08/will-graphene-replace-silicon-in.html

 

[Rocketman]

As far as I can tell, graphene will be the next great enabler, but it's still quite a ways off.

Been following that. I agree that and q-bits are the next thing. But as I have been saying for a few years now computer chips right now are "good enough" for 98% of uses and improved I/O is the bottleneck. TB and SSD puts a big dent in that and has legs for future improvement.

People with heavy processing requirements can use blades (stacked boards) and grids (stacked computers) or servers (big iron).

I think the % of folks who really need a computer faster than a 2013 Mac is so tiny, the addressable market is best served by VARs.

Rocketman

 

[chrmjenkins]

Been following that. I agree that and q-bits are the next thing. But as I have been saying for a few years now computer chips right now are "good enough" for 98% of uses and improved I/O is the bottleneck. TB and SSD puts a big dent in that and has legs for future improvement.

People with heavy processing requirements can use blades (stacked boards) and grids (stacked computers) or servers (big iron).

I think the % of folks who really need a computer faster than a 2013 Mac is so tiny, the addressable market is best served by VARs.

Rocketman

I/Os will be taken care of by hybrid transistors and other optical interconnects realizable on silicon substrates. That's where TB is headed with Light Peak.

 

[jonnysods]

Maybe they are getting ready to use this in other hardware?

 

[topicolo]

But only once before. Apple has been very aggressive with their SoC plans, actually. Looking at this move, I don't think the AppleTV alone has the volume to justify a die-shrunk A5. So, you have two scenarios:

1) Shrunk A5 finds it way into 4S to cut costs and improve battery life. Next iPhone gets new chip.
2) Shrunk A5 finds it way into 4S to cut costs and improve battery life. Next iPhone gets same A5.

2) is very possible, but given Apple's aggressive track record, I'd imagine they would want to keep pushing the envelope and do something new. iOS 6 unveil will be a huge hint as to what to expect. If it does a lot of new, compute heavy things, a new CPU seems more likely to me. Of course, a lot of those special features may not be unveiled until the new iPhone is, as they would remain exclusive to it.



No, it would require a firmware (BIOS) hack, if the core wasn't hard disabled or non-functional.

Very good points and good logic. The die shrink seems to be powering the iPad 2 as well, so the combination of TV, iPad 2, iPod and possible smaller iPad may be enough of a market for Apple. I do think that the current A5x as it stands will be far too power hungry for the iPhone but die shrinks and whatnot can change things in the next 5-6 months

 

[chrmjenkins]

Very good points and good logic. The die shrink seems to be powering the iPad 2 as well, so the combination of TV, iPad 2, iPod and possible smaller iPad may be enough of a market for Apple. I do think that the current A5x as it stands will be far too power hungry for the iPhone but die shrinks and whatnot can change things in the next 5-6 months

The A5X is a bruteforce stop-gap solution. I wouldn't be surprised if we never see it in a product again, even with a die shrink. By the time of the next iPad, they can have a SoC out with a dual-core Rogue solution that will be even faster than the 543MP4.

 

[Eidorian]

Interesting chart. This indicates to me the current chip release may have particularly long legs, perhaps 3 years or more. 2012-13, 2013-14, 2014-15 and probably into 2015-16 too. Getting people used to that idea would reduce the whining.

Rocketman

It is a mess that AMD wanted to get out of. Intel can afford its own fabs and to retool them for a new node but everywhere else everyone shares the pain of a new node.

There is talk that Haswell's Lynx Point PCH is going to have a lower TDP but I just suspect that the PCH is just moving to 45nm or even 32nm. Intel was using 65nm for PCHs but that information gets murky beyond the 5 Series. AMD still relies on 65/40nm bulk for everything that is not a CPU/APU/GPU. The original Atom suffered from the 945GC* as the chipset in a whopping 90nm. The northbridge ended up consuming more power and being larger than the CPU itself. Even the 4 + 1 Terga 3 is in 40nm.

I am sure people wish everything on the board was made in the lowest process available.

 

[MagnusVonMagnum]

Too bad the new AppleTV isn't dual-core enabled. Maybe it could have supported 4.1 compression instead of just 4.0, not to mention more potential for future gaming if they ever allow app sales for it.