Apple Shifting to TSMC for A-Series Chip Production Earlier Than Expected?

[blackhand1001]

Originally Posted by chrmjenkins

The exynos is largely a stock A15 implementation. There is much more design in the Swift A6 core than there is the exynos 5 core. Most of Samsung's work went into the memory hierarchy.

That being said, there's nothing spectacular about Samsung's designs. In fact, most of the design wins go to Nvidia, who can leverage their own graphics IP and have been forward thinking with their shadow core, or Qualcoomm, who has been doing custom ARM architecture implementations long before apple did swift.

Samsung has also yet to fit Exynos 5 in a phone or even a phablet. Thus, it's clear it wouldn't have been ready for the iphone 5 and it hasn't proven itself as a phone processor yet at all.


Although I can find no literature on the die size, you're looking at the issue wrong. Power is a combination of process, transistors, transistor type, operating frequency, dynamic operating frequency, core voltage and power saving implementations that can be transistor level logic implementations or endemic like power gating and declocking. It also depends heavily on what fills the area such as cache, core logic, memory bus, etc. Different areas see different toggle rates, and hence, more power draw.



A15 is no doubt more powerful than the A9 and recent custom implementations of the ARMv7s ISA, but they're starting to add a lot of fluff phone processors don't need. ARM is looking to take on Intel in the ultrabook/notebook and eventually even server space with their recent 64-bit processor announcements. Not really what Apple needs in a phone processor. ARM's introduction of the light A7 core is in fact a reflection of the fact that can't do a one size fits all approach with the A15.

It's also important to remember, as was noted, that their clocks are 25% higher.



The iPad needs an X processor because it has more pixels to drive. The touch has an older processor because it's inherently a much more low margin device. Also why it has a worse camera and a worse screen.



TSMC doesn't have a 32nm node. They skipped it. But yes, generally "half-nodes" do retain rules and a simple optical shrink is usually doable.

The exynos 5250 is only clocked 20% higher than the a6x yet performs 70% better. That also means that it does much better on instructions per clock. The a6 architecture is somewhere between a9 and a15 much like Qualcomm krait is. Both are vastly inferior to the a15 core. The exynos 5250 also has excellent power management. The a6x needs a gigantic battery in the iPad. The nexus 10 has a smaller battery yet gets about the same battery even when pushing more pixels.

[KdParker]

Originally Posted by NakedPaulToast

This post reminds of all the "Bye-Bye, Google " posts when iOS went to Apple Maps.

How'd that work out?

Pretty good for most of us in the US.

But having a new chip is easy enough to test vs the A series to make sure that there is not any drop off.

[SockRolid]

Originally Posted by MacRumors

... although it has indicated that it is open to dedicating a factory or two to a single customer.

Tim Cook: "Hello TSMC. This is Tim."

TSMC: "Hello, Mr. Cook. We've been looking forward to talking with you."

Tim Cook: "I'm going to wire you 2 billion dollars. Build us a factory or two for A7 chip production."

TSMC: "Yes sir. When would you like full-scale production to begin?"

Tim Cook: "How 'bout 28 nano production in late 2013, 20 nano production in early 2014?"

TSMC: "That should be as easy as cake."

Tim Cook: "Pie. As easy as pie."

TSMC: "Yes sir. Easy as pie, sir."

Tim Cook: "And you can just call me Tim."

TSMC: "Yes sir, Mr. Tim."

[chrmjenkins]

Originally Posted by blackhand1001

The exynos 5250 is only clocked 20% higher than the a6x yet performs 70% better. That also means that it does much better on instructions per clock. The a6 architecture is somewhere between a9 and a15 much like Qualcomm krait is.

Krait is around 3.2 DMIPS. A15 is 3.5. I assume Swift is near what Krait is. The true metric for a mobile device is performance per watt, where Krait and Swift do just fine.

Both are vastly inferior to the a15 core. The exynos 5250 also has excellent power management. The a6x needs a gigantic battery in the iPad. The nexus 10 has a smaller battery yet gets about the same battery even when pushing more pixels.

Ehhh, yes they have a lower IPC, but that doesn't really tell the story. If qualcomm and apple had thought a vanilla core would have been better than what they could design, they would have gone with it. For instance, I can tell you that Swift and Krait inherently handle pipeline flushes better by virtue of having 11 and 10 stages respectively versus 15 on the A15. I'm sure there are plenty of more optimizations and deviations from standard library cells to squeeze every last performance per watt out of it.

The Nexus 10 drives marginally more pixels. The A6X has a much higher capability GPU, which is what truly drives its battery size.

[KdParker]

Originally Posted by HobeSoundDarryl

Stand by. That "no reason" is gaining an understanding of how to build these for Apple and watching Apple's revenues grow and grow. If I baked cakes for your bakery and watched you grow richer and richer on my baking, it's not long before I start thinking about selling my own cakes direct.

Having watched this play out over and over in my life, I would bet heavily that it's only going to be a matter of time before TSMC begins to expand into businesses beyond just fabbing.

And even if TSMC expands thier business, It doesn't mean it will be another Andriod phone. Who know how this could play out....

IOS/Windows phones with different form factors??? Who knows what will be next.

[blackhand1001]

Originally Posted by chrmjenkins

Krait is around 3.2 DMIPS. A15 is 3.5. I assume Swift is near what Krait is. The true metric for a mobile device is performance per watt, where Krait and Swift do just fine.



Ehhh, yes they have a lower IPC, but that doesn't really tell the story. If qualcomm and apple had thought a vanilla core would have been better than what they could design, they would have gone with it. For instance, I can tell you that Swift and Krait inherently handle pipeline flushes better by virtue of having 11 and 10 stages respectively versus 15 on the A15. I'm sure there are plenty of more optimizations and deviations from standard library cells to squeeze every last performance per watt out of it.

The Nexus 10 drives marginally more pixels. The A6X has a much higher capability GPU, which is what truly drives its battery size.

Hardly marginally more pixels, the nexus has 4,096,000 while the ipad has 3,145,728. Thats nearly a million pixel difference. The exynos cpus have always had great power management. The fact is the exynos cortex a15 outperforms the a6 by a very large margin, and it also clocks much higher with similar power usage. The quad core 5450 version with the Mali-658 is imminent which ups the game even more so. It will be clocked at 2ghz.

[tbrinkma]

Originally Posted by chrmjenkins

The iPad needs an X processor because it has more pixels to drive. The touch has an older processor because it's inherently a much more low margin device. Also why it has a worse camera and a worse screen.

The iPod Touch can also use an older processor because it doesn't have to run all of the phone-related background processes that the iPhone needs.

[chrmjenkins]

Originally Posted by blackhand1001

Hardly marginally more pixels, the nexus has 4,096,000 while the ipad has 3,145,728. Thats nearly a million pixel difference. The exynos cpus have always had great power management. The fact is the exynos cortex a15 outperforms the a6 by a very large margin, and it also clocks much higher with similar power usage. The quad core 5450 version with the Mali-658 is imminent which ups the game even more so. It will be clocked at 2ghz.

There's nothing wrong with either of their power management performance. That's not really in question here. Performance per watt matters and that's why the iphone 5 with a 50% or more smaller battery still has comparable battery life to android counterparts.

And I've told you clocks don't matter. Intel spent the good part of the early 2000's chasing MHz with NetBurst and AMD handed their ass to them when they went 64-bit and dual core first. It matters how efficient and smart your architecture is, not how high you can clock it. The clock speeds on Krait and Swift are going to be inherently lower because they have fewer pipeline stages. That means you have more functional logic in each stage and can't push through an individual stage as fast.

[firewood]

Originally Posted by SockRolid

"I'm going to wire you 2 billion dollars. Build us a factory or two for A7 chip production."

Sounds like you haven't kept up with the costs of building a new semiconductor/ASIC fab line of recent.

See the wikipedia entry on Rock's law and note that the 2003 price given is almost a decade obsolete (e.g. tiny by todays standards).

[Lesser Evets]

Seems to fit into the existing rumors.

iPhone 5s in late Spring 2013 or even early summer.

iPhone 6 in the end of 2013.

With these releases, Apple spanks the competition out of the field in all ways except one: price. The rumored slide of Apple in the iDevices market would reverse completely and they could trounce everyone within the decade.

[SockRolid]

Originally Posted by firewood

Sounds like you haven't kept up with the costs of building a new semiconductor/ASIC fab line of recent.

See the wikipedia entry on Rock's law and note that the 2003 price given is almost a decade obsolete (e.g. tiny by todays standards).

Tim Cook: "Here's $2 billion up front. Build us a factory or two for A7 chip production. Let me know what the total cost will be."

[blackhand1001]

Originally Posted by chrmjenkins

There's nothing wrong with either of their power management performance. That's not really in question here. Performance per watt matters and that's why the iphone 5 with a 50% or more smaller battery still has comparable battery life to android counterparts.

And I've told you clocks don't matter. Intel spent the good part of the early 2000's chasing MHz with NetBurst and AMD handed their ass to them when they went 64-bit and dual core first. It matters how efficient and smart your architecture is, not how high you can clock it. The clock speeds on Krait and Swift are going to be inherently lower because they have fewer pipeline stages. That means you have more functional logic in each stage and can't push through an individual stage as fast.

Comparing it to netburst is a bad comparison as cortex a15 not only clocks higher, it has a better instructions per clock as well. Thats much different than the athlon x64 vs pentium 4 where the pentium 4 clocked higher but the athlon 64 had much better ipc.

[chrmjenkins]

Originally Posted by blackhand1001

Comparing it to netburst is a bad comparison as cortex a15 not only clocks higher, it has a better instructions per clock as well. Thats much different than the athlon x64 vs pentium 4 where the pentium 4 clocked higher but the athlon 64 had much better ipc.

That's why my point was focusing on MHz is not a good idea, and nothing more.

The athlon had better IPC because they made architecture optimizations and advancements. That's why Apple and Qualcomm pursue custom architectures. Lower IPC withstanding, it still must have yielded a net benefit for their typical workloads else they wouldn't have invested millions and millions in a custom design when vanilla would have done.

[firewood]

Originally Posted by daneoni

A newer processor is faster than an older one?

Exactly. Look at the date each processor was available on a pocketable cell phone that you could actually order, and compare those dates of availability with what should be expected according to Moore's law. Nothing of significant interest here.

In addition, according to ARM's own architecture papers, the A15 is optimized for a very different instruction mix and application load. Until lots of apps needing that type of instruction mix are available, an A15 implementation will likely be non-optimal for mobile use.

[blackhand1001]

Originally Posted by chrmjenkins

That's why my point was focusing on MHz is not a good idea, and nothing more.

The athlon had better IPC because they made architecture optimizations and advancements. That's why Apple and Qualcomm pursue custom architectures. Lower IPC withstanding, it still must have yielded a net benefit for their typical workloads else they wouldn't have invested millions and millions in a custom design when vanilla would have done.

Except the fact that the exynos not only can run at a higher clockrate but even at the same clock speed as the a6 it outperforms it.

[thekev]

Originally Posted by SeattleMoose

Good to see Apple trying to move away from Samsung components. Reduces leverage/control over Apple products. Of course Samsung is such an honorable company and would NEVER claim "part shortage" to hurt Apple..

You guys make too much stuff up. What makes you think these companies would ignore their contracts?

Originally Posted by HobeSoundDarryl

How long until TSMC decides to do a Samsung on Apple? In other words, after making these for Apple for a while, how long until TSMC starts thinking: why don't we make our own phones, tablets, pods, etc? It worked so well for Samsung; it could work for us too (Apple has shown us the way).

I see many of us are celebrating Apple's "victory" in further moving away from Samsung. Why we can't look forward and see that Apple is probably just creating another Samsung (or three) is beyond me.

This is too ignorant for words. Go look up the history of both Samsung and LG. They were both making phones long before Apple. You may not like their phones, but that is really irrelevant.

[blackhand1001]

Originally Posted by firewood

Exactly. Look at the date each processor was available on a pocketable cell phone that you could actually order, and compare those dates of availability with what should be expected according to Moore's law. Nothing of significant interest here.

In addition, according to ARM's own architecture papers, the A15 is optimized for a very different instruction mix and application load. Until lots of apps needing that type of instruction mix are available, an A15 implementation will likely be non-optimal for mobile use.

the exynos 5250 shipped in a device a month after the a6 shipped and before the ipad 4 was even announced let alone shipped. Its barely any newer.

[Glideslope]

Yea Baby!!!!!!!!!!!!!!!!!!!!!!!!!!

----------

Originally Posted by NakedPaulToast

This post reminds of all the "Bye-Bye, Google " posts when iOS went to Apple Maps.

How'd that work out?

Uh, Bob Mansfield is not Scotty Boy. TSC is basically Apple's own Fab unit. By the time Samsung's Contract is up in 6-14, Apple will be well positioned. They have invested over 3 Billion in the Fab. Soon N & S Korea will reunite. Probably best to not be Korean Heavy at that point. Expect LG displays to go after Samsung.

[chrmjenkins]

Originally Posted by blackhand1001

Except the fact that the exynos not only can run at a higher clockrate but even at the same clock speed as the a6 it outperforms it.

Apparently you haven't been reading. Not only have I acknowledged a higher IPC, I said the actual rates:

Originally Posted by chrmjenkins

Krait is around 3.2 DMIPS. A15 is 3.5. I assume Swift is near what Krait is. The true metric for a mobile device is performance per watt, where Krait and Swift do just fine.

Originally Posted by blackhand1001

the exynos 5250 shipped in a device a month after the a6 shipped and before the ipad 4 was even announced let alone shipped. Its barely any newer.

In a much lower volume device. They have to make chips months in advance to meet launch demand. They would need less lead time for a smaller quantity.

[samcraig]

Poor poor early adopters/beta testers...

[paulrbeers]

Originally Posted by chrmjenkins

Apparently you haven't been reading. Not only have I acknowledged a higher IPC, I said the actual rates:





In a much lower volume device. They have to make chips months in advance to meet launch demand. They would need less lead time for a smaller quantity.

Don't feed the android trolls! They don't bother to read/comprehend valid arguments.

Fact of the matter is, at some point our phones have enough power to do what is needed of them and whether you run a cortex A9, Swift, Krait, or A15, it really doesn't matter. Especially since iOS is more GPU heavy, it matters even less. Frankly why would Apple have spent a fortune on ARM design firms and NOT build their own ARM processors? It just doesn't make sense. Swift processors are really impressive for what they are and the amount of power they use. It is more than enough to stay competitive with the A15 crowd for this generation. Most likely the A7's will be A15 derivatives.

[blackhand1001]

Originally Posted by Glideslope

Yea Baby!!!!!!!!!!!!!!!!!!!!!!!!!!

----------



Uh, Bob Mansfield is not Scotty Boy. TSC is basically Apple's own Fab unit. By the time Samsung's Contract is up in 6-14, Apple will be well positioned. They have invested over 3 Billion in the Fab. Soon N & S Korea will reunite. Probably best to not be Korean Heavy at that point. Expect LG displays to go after Samsung.

You really think apple really could actually meet demand while not using Samsung or lg displays. They are the two largest producers.

[Verbatim Cookie]

Originally Posted by glideslope

Soon N & S Korea will reunite.

WAT Just some phantom text so that capital letters appear above.

[fpsBeaTt]

Originally Posted by NakedPaulToast

This post reminds of all the "Bye-Bye, Google " posts when iOS went to Apple Maps.

How'd that work out?

I honestly never saw any quote remotely like that, at least none implying that Apple's switch would signify 'the end of Google'.

[WatchTheThrone]

Originally Posted by blackhand1001

Except the fact that the exynos not only can run at a higher clockrate but even at the same clock speed as the a6 it outperforms it.

You missed the android forum by about 10 exits!!!!