[WSJ] TSMC is shipping 20nm SOCs to Apple

[raghu78]

desktop class? You're swimming in apple's kool-aid I guess simply drinking too much of it is not enough. Desktop? Really? They would need to triple the frequency or double the frequency and the number of cores for that to be true but only for CPU part. I don't think its GPU part compares favorably to crystalwell or 512 GCN cores.

http://www.anandtech.com/show/7910/apples-cyclone-microarchitecture-detailed

" With six decoders and nine ports to execution units, Cyclone is big. As I mentioned before, it's bigger than anything else that goes in a phone. Apple didn't build a Krait/Silvermont competitor, it built something much closer to Intel's big cores. At the launch of the iPhone 5s, Apple referred to the A7 as being "desktop class" - it turns out that wasn't an exaggeration."

btw Crystalwell takes 264 sq mm on Intel 22nm FINFET. A7 is 102 sq mm on Samsung 28nm. Given the same die size Apple can do a much better job than Intel graphics with PowerVR Rogue 6XT.

Apple uses the A7 in tablets and phones. Can Apple scale the Cyclone core to run at higher frequencies. Yeah they can if they wanted to. Obviously that means it will draw more power. If Intel can scale Broadwell from 4.5w dual core tablet skus to high end server CPUs with 18 cores and 140w TDP, Apple definitely can. Apple is waiting for FINFET to make a wider transition to ARMv8. Also Apple needs to get the software infrastructure ready. that takes time.

 

[Nothingness]

Apple is waiting for FINFET to make a transition to ARMv8.

Apple A7 CPU already is ARMv8.

 

[coercitiv]

Found Intel's comparable setup to the Apple A7.

i3-4010Y / 11.5W TDP / 6W SDP
http://browser.primatelabs.com/geekbench3/401702

Apple A7 / 2W TDP / <1W SDP
http://browser.primatelabs.com/geekbench3/585335

Interesting. Let's follow your line of thought and test things at 1-2W power usage, we might end up uncovering something:

Intel Atom Z3770 / 2W SDP
http://browser.primatelabs.com/geekbench3/497549

It beats both A7 and Haswell in multithreaded scenarios. Based on these benchmark scores, would you make the following statement:

"Silvermont is much more efficient and much more faster than Haswell."

 

[ShintaiDK]

btw Crystalwell takes 264 sq mm on Intel 22nm FINFET. A7 is 100 sqmm. Given the same die size Apple can do a much better job than Intel graphics with PowerVR Rogue 6XT.

Crystalwell is the eDRAM. I assume you wanted to show the size of a Haswell quadcore with GT3.

You also compare a dualcore with a quadcore.

 

[raghu78]

Apple A7 CPU already is ARMv8.

I meant a wider transition across the product stack which includes the macbook air / pro which are x86-64 currently and running Intel chips.

Crystalwell is the eDRAM. I assume you wanted to show the size of a Haswell quadcore with GT3.

You also compare a dualcore with a quadcore.

I am saying if Apple decides to design ARMv8 based SOCs which serve higher up the stack they can do a better job than Intel especially in graphics. CPU wise too they can build something which competes with Intel's best at same die size and TDP, especially at TSMC 16FF+ / Samsung 14LPE. We will see by the time A9 ships if Apple has plans to use Ax in other products.

 

[ShintaiDK]

Apple A7 / 2W TDP / <1W SDP.

The A7 consume much more than that

 

[ShintaiDK]

I meant a wider transition across the product stack which includes the macbook air / pro which are x86-64 currently and running Intel chips.

Even if we somehow imagine the performance would be there. Then sales of Macbook Air and Macbook Pro would plummet. Since a large part of the buyer base actually install windows. It would also pretty much eliminate Apples last niche in the business sector. And it would destroy the compability with all current apps and games for OSX.

 

[Exophase]

It's only been 8 years since the Apple used the PowerPC 970.

Yeah, nothing new for Apple at any rate. 8 years is still pretty long but I guess it felt like it was even longer ago.. and maybe I didn't weigh it as heavily in my mind since so fewer people used it (vs an iOS CPU) and the machines were at least less restricted for putting other OSes on it.

I'm just disappointed that no one's been able to port Linux to any A7-based devices. Probably has something to do with having a locked bootloader. But then, Linux support isn't all that great on the devices where the bootloader has been cracked.

You're right, Linux support really isn't that good for most mobile SoCs. But at least you see some one off products that try to make a go of it.

Because it will be substantially worse than competing SoCs (Core M).

Okay, so I guess from your posts here you think that Intel Core processors will be in phones soon...

 

[witeken]

I meant a wider transition across the product stack which includes the macbook air / pro which are x86-64 currently and running Intel chips.

That would be a downgrade.

I am saying if Apple decides to design ARMv8 based SOCs which serve higher up the stack they can do a better job than Intel especially in graphics. CPU wise too they can build something which competes with Intel's best at same die size and TDP, especially at TSMC 16FF+ / Samsung 14LPE. We will see by the time A9 ships if Apple has plans to use Ax in other products.

We haven't seen any proof of anything you say; Core M vs A8 will be an interesting comparison.

 

[witeken]

Okay, so I guess from your posts here you think that Intel Core processors will be in phones soon...

There is also the iPad Air. Apple convinced Intel to invest in graphics, so if they ask Intel, they would maybe make a phone SoC with Core.

 

[ancientarcher]

There is also the iPad Air. Apple convinced Intel to invest in graphics, so if they ask Intel, they would maybe make a phone SoC with Core.

and sell a phone with a fan attached, I guess :biggrin:
iFan anyone?

 

[Exophase]

There is also the iPad Air. Apple convinced Intel to invest in graphics, so if they ask Intel, they would maybe make a phone SoC with Core.

So I guess this is Apple's timeline:

1) Invest a ton of money buying PASemi and Intrinsity to get high quality uarch and layout engineers and IP
2) Develop cutting edge 64-bit ARMv8 CPU for their highly custom SoC, with feature set and release schedule aligned to the rest of their product design
3) Throw it all out and ask Intel to put their desktop chips on a phone for them despite Intel showing no inclination to do so - because Intel has better transistors (like they have for a long time)

Sounds like a winner. You should already know this, but Apple convincing Intel to release SKUs with higher end graphics for product lines they're already using doesn't come anywhere close to this. It'd be a huge deal for Intel and something that Apple is probably not remotely interested in.

 

[Khato]

Found Intel's comparable setup to the Apple A7.

i3-4010Y / 11.5W TDP / 6W SDP
http://browser.primatelabs.com/geekbench3/401702
(heh, heh, model 69)

Apple A7 / 2W TDP / <1W SDP
http://browser.primatelabs.com/geekbench3/585335

Even disregarding potential issues with cross-platform geekbench comparisons... where exactly did you come up with the TDP and SDP numbers for the A7? Out of thin air I presume?

 

[kimmel]

and sell a phone with a fan attached, I guess :biggrin:
iFan anyone?

2010 called, they want their joke back.

Even disregarding potential issues with cross-platform geekbench comparisons... where exactly did you come up with the TDP and SDP numbers for the A7? Out of thin air I presume?

Much like 98% of phone statistics. They are made up by forum users who try to sound like they know what they are talking about.

 

[witeken]

and sell a phone with a fan attached, I guess :biggrin:
iFan anyone?

You don't need a fan with a 4.5W SoC, which is as much as A7. Have you seen Intel's reference Core M tablet? It's thinner than the iPad Air.

So I guess this is Apple's timeline:

1) Invest a ton of money buying PASemi and Intrinsity to get high quality uarch and layout engineers and IP
2) Develop cutting edge 64-bit ARMv8 CPU for their highly custom SoC, with feature set and release schedule aligned to the rest of their product design
3) Throw it all out and ask Intel to put their desktop chips on a phone for them despite Intel showing no inclination to do so - because Intel has better transistors (like they have for a long time)

Sounds like a winner. You should already know this, but Apple convincing Intel to release SKUs with higher end graphics for product lines they're already using doesn't come anywhere close to this. It'd be a huge deal for Intel and something that Apple is probably not remotely interested in.

I don't expect Apple to switch to Intel in the near future. But you can have the best architecture in the world (which is difficult with Intel as competitor), if it has to run on 5 year old transistors, it won't be able to compete. So when Intel releases its 10nm with III-V transistors in 2016, Apple can only use 20nm FinFETs which Intel has been shipping since 2012. If Apple wants III-V from TSMC, they'll have to wait until ~2021.

Then it does become a legit question: does Apples keep designing its own architecture or do they switch to Intel's chips on a far superior process node? If III-V is as good as it is rumored to be (up to an order of magnitude lower power consumption), then I see no reason why Apple would not want to switch to Intel in 2016.

 

[AtenRa]

that's wrong math, using your math there would be no penalty to having a big die instead of a small one. Just use a hypothetical 650mm2 chip and do the math. 92* dice per wafer with perfect yields and with 60% yields you would get 55 functional dice, that's just wrong.

that's assuming your math is correct for the number of 100mm2 dice in a 300mm wafer.

I took 60% yields for the 100mm^2 dies, if you have a 650mm^2 die then yields would not be 60% but 20% or lower.

 

[NTMBK]

Even if we somehow imagine the performance would be there. Then sales of Macbook Air and Macbook Pro would plummet. Since a large part of the buyer base actually install windows. It would also pretty much eliminate Apples last niche in the business sector.

Do you have any statistics to back that up? I've seen no evidence that a "large part" of the Mac user base buy and install Windows.

And besides, there's always Windows RT

And it would destroy the compability with all current apps and games for OSX.

They already killed backwards compatibility when they moved from PowerPC to x86.

 

[Homeles]

Bah, why isn't anybody excited at what the A8 will bring?

The A7 was really neat, so personally, I can't wait to see the A8 in all of its 20-nanometer glory. Let's see what Apple/TSMC can do. I'm expecting that both companies will execute well.

Even if Apple did nothing else on the CPU side, other than moving to a quad core, the A8 would be pretty crazy. It'll be interesting to see what they've gained from throwing their wallet at TSMC.

Do you have any statistics to back that up? I've seen no evidence that a "large part" of the Mac user base buy and install Windows.

I don't know about "large," but it's not an insignificant number. They're pretty popular at universities, where that kind of compatibility is often needed.

 

[Roland00Address]

that's wrong math, using your math there would be no penalty to having a big die instead of a small one. Just use a hypothetical 650mm2 chip and do the math. 92* dice per wafer with perfect yields and with 60% yields you would get 55 functional dice, that's just wrong.

that's assuming your math is correct for the number of 100mm2 dice in a 300mm wafer.

If the errors per die are equally distributed then you square the percent each time you double the length of a chip (assuming a square chip for simplicity).

So lets assume 60% successful chip for a 100mm2 chip (10mm x 10 mm). You can get about 600 dies out of a 300 mm2 wafer with 100% yield. So we are talking about 360 successful dies (60%).

Make the chip 400mm2 we are talking about 140 dies per wafer with 100% yields, with 60% yeilds at 100mm2 we are talking, if 60%*60%=36% or only 50 successful dies with a 400mm2 chip.

Make the chip 625mm2 (number picked for easy math and you have an error rate of 60%^2.5 power or 27.8% functional functional units. With 100% yields you would have 89 dies per wafer, but we are talking about only about 24 or so successful dies.

 

[jhu]

I don't get this. I thought "CPUs and Overclocking" was about people buying processors for computers they made, and maybe for those that OC them. I see no one in the wild being able to buy one of these CPUs for their computer, let alone OC.

Should move to the "All Things Apple" forum.

Its "CPUs and Overclocking" for a reason and not "CPUs for the computer build I bought to overclock". That's why people talk about Xeons, Opterons, Xeon Phi, POWER, etc. here and not just the latest from Intel that can be bought at Microcenter.

 

[Lepton87]

I took 60% yields for the 100mm^2 dies, if you have a 650mm^2 die then yields would not be 60% but 20% or lower.

NVM

 

[ams23]

Even if Apple did nothing else on the CPU side, other than moving to a quad core, the A8 would be pretty crazy. It'll be interesting to see what they've gained from throwing their wallet at TSMC.

The expectation is that Apple will stick with a dual core Cyclone-derived CPU in the iPhone 6 (with max CPU clock operating frequencies close to ~ 2 GHz), while the GPU will likely be the G6630 or GX6650 (with max GPU clock operating frequencies close to ~ 600 MHz).

 

[Exophase]

You don't need a fan with a 4.5W SoC, which is as much as A7. Have you seen Intel's reference Core M tablet? It's thinner than the iPad Air.

It also has a 12.5" display, now why do you think that is? Kind of means there's more to that comparison than thickness. And why look at iPad Air thickness when we're talking about phones, which A7 is also in? If you really think that CPU can fit okay in a phone I've got news for you.

I don't expect Apple to switch to Intel in the near future. But you can have the best architecture in the world (which is difficult with Intel as competitor), if it has to run on 5 year old transistors, it won't be able to compete. So when Intel releases its 10nm with III-V transistors in 2016, Apple can only use 20nm FinFETs which Intel has been shipping since 2012. If Apple wants III-V from TSMC, they'll have to wait until ~2021.

Then it does become a legit question: does Apples keep designing its own architecture or do they switch to Intel's chips on a far superior process node? If III-V is as good as it is rumored to be (up to an order of magnitude lower power consumption), then I see no reason why Apple would not want to switch to Intel in 2016.

Lately Intel has been promoting both their 14nm and 10nm nodes quite a bit. You'd think somewhere in the slides they would have included that 10nm will also have an order of magnitude power consumption improvement. So I don't know why you think Intel's 10nm will be made using III-V materials, other than that they were hopeful about it maybe 5 years ago. Actually, your comments about power consumption also sounds like things they were hopeful about that long ago. But it seems pretty clear to me that III-V is not on the table for 10nm anymore, eg http://www.eetimes.com/document.asp?doc_id=1263255

Here's another good reason for Apple not wanting to use Intel chips that I didn't list in the post above - the big ARM software library and the work they've put in to support it, including the recent transition to ARMv8...

 

[TuxDave]

Geekbench is actually pretty good as long as you're looking at 3.1 scores.

Except Geekbench 3.1 is still too lazy to optimize traces that have been trivially optimized up the wazoo by everyone else:

Processor Intel Core i3-4010Y @ 1.30 GHz

DGEMM
Single-core 1168
1.72 Gflops

Unless they're trying to trivially cause cache misses (which makes the allocation/decode/execution width kind of pointless), at 1.3GHz you should be seeing 10x the performance since AVX2 is trivially useful for DGEMM.

http://browser.primatelabs.com/geekbench3/401702

 

[witeken]

It also has a 12.5" display, now why do you think that is? Kind of means there's more to that comparison than thickness. And why look at iPad Air thickness when we're talking about phones, which A7 is also in? If you really think that CPU can fit okay in a phone I've got news for you.

What do you mean? Microsoft's Surface Pro 3 is also 12". Phones? Sure, Llama Mountain is 7.2mm thick, less than iPhones 5s' 7.6mm. Core M isn't made for phones, though.

Lately Intel has been promoting both their 14nm and 10nm nodes quite a bit.

Intel has only been promoting 14nm. They only mentioned 10nm for their cost/transistor slides.

You'd think somewhere in the slides they would have included that 10nm will also have an order of magnitude power consumption improvement.

14nm products haven't even launched. They'll talk about 10nm when it's appropriate. I'm also not saying power consumption will be an order of magnitude less, but from what I've read it's possible, depending on the materials. My point was that when we're talking about an order of magnitude with those materials, Intel could take a massive lead that would give them an unambiguous leadership position.

So I don't know why you think Intel's 10nm will be made using III-V materials, other than that they were hopeful about it maybe 5 years ago. Actually, your comments about power consumption also sounds like things they were hopeful about that long ago. But it seems pretty clear to me that III-V is not on the table for 10nm anymore, eg http://www.eetimes.com/document.asp?doc_id=1263255

That link doesn't say that 10nm won't use III-V. First it seemed that 10nm would use III-V, but then I saw this roadmap from Applied Materials, saying Intel will use Germanium. But those 2 slides, from ASML and Intel, say 10nm will use III-V and Germanium.

http://download.intel.com/newsroom/...esearch_Enabling_Breakthroughs_Technology.pdf
http://www.sokudo.com/event/images/130710/5_ASML.pdf