Anand: Apple's A7 Cyclone Microarchitecture Detailed (2014-03-31)

[Eug]

Apple's Cyclone Microarchitecture Detailed

---

It wasn't until I wrote this piece that Apple's codenames started to make sense. Swift was quick, but Cyclone really does stir everything up.

Looking at Cyclone makes one thing very clear: the rest of the players in the ultra mobile CPU space didn't aim high enough. I wonder what happens next round.

---

Much of the article is over my head, but nonetheless it seems like A7's arch may be around for some time, since much of its power has not been leveraged yet.

I hope to see a 2 GB iPad Air this year. That's what I'd like to buy. I was fine to get an iPhone 5s with 1 GB, but I tend to keep my tablets longer than I keep my phones.

 

[saratoga172]

I think the likely next jump is can they and how do they integrate this into a tablet/computer hybrid. Will we see this running a version of OS X? For example.

 

[Homeles]

The ROB is just stupidly large.

 

[Rakehellion]

I think the likely next jump is can they and how do they integrate this into a tablet/computer hybrid. Will we see this running a version of OS X? For example.

It's fast for mobile, but even an i5 trumps it by a wide margin.

 

[blackened23]

It's fast for mobile, but even an i5 trumps it by a wide margin.

Well, of course. At this time, yes, the i5 is better. But if apple can do i'm sure they would want to. It just not quite there yet.

His statement was more about the future, who knows what could happen there. Certainly no one is close to intel's core architecture at this point, but if it happens 5 years from now - hey - more competition and all that sort of thing. Intel is in a race to get their core architecture at even lower power points and they're doing amazingly well at that with their -Y mobile SKUs, while ARM SOC vendors are in a race to increase performance. Where this ends up will be interesting for sure, IMO. That said, I don't see mobile SOCs over-taking core anytime even remotely soon, not even in the next 5-7 years. The performance lead intel has even in their mobile ultra low voltage SKUs is just tremendous; while intel is pretty darn close in terms of efficiency. Probably too much to overcome in even several years time. If i'm wrong though, that's cool, it'll just drive the competition stakes up higher which will be fun to watch.

 

[xpenryn]

Anand said he had confirmed that Cyclone had 6 decoders, but the llvm document doesn't have any info about decoder's amout. Although Anand did some test, i saw some comments criticized his test method. One expert from China ICCT told me that there's no program could calculate the amount of decoders, anand might made an amateur mistake.He also said
that it was impossible for apple to develop a microarchitecture which had 6 decoders, for it will be too complicated.

 

[Lil Frier]

I think the likely next jump is can they and how do they integrate this into a tablet/computer hybrid. Will we see this running a version of OS X? For example.

If people complain that Windows 8 is a bastardization of two UIs, why would folks consider an iOS/OSX hybrid to be any different/better?

 

[saratoga172]

It's fast for mobile, but even an i5 trumps it by a wide margin.

I agree they are no where close to intels core architecture in terms of performance. But the A7 was less about taking a claim right now and more about setting in the seeds of the future.

If they can continue to optimize and improve upon what they have they could be say where AMD is in 5 years with better efficiency. But have less worry being that they would be developing for their own hardware/software. I see the bigger challenge in getting devs to change architecture from an x86 instruction set to the arm based or whatever A7 utilizes. But that's sort of why I mentioned some form of cross over device. My thinking is less about the power segment (design studios) and more about mainstream where pure power is not required. People want a smooth experience.

Little off topic but I get where you're going. I agree but think apple would choose to target a different segment with the product. One I think would fill a large majority of the need and want.

 

[saratoga172]

Well, of course. At this time, yes, the i5 is better. But if apple can do i'm sure they would want to. It just not quite there yet.

His statement was more about the future, who knows what could happen there. Certainly no one is close to intel's core architecture at this point, but if it happens 5 years from now - hey - more competition and all that sort of thing. Intel is in a race to get their core architecture at even lower power points and they're doing amazingly well at that with their -Y mobile SKUs, while ARM SOC vendors are in a race to increase performance. Where this ends up will be interesting for sure, IMO. That said, I don't see mobile SOCs over-taking core anytime even remotely soon, not even in the next 5-7 years. The performance lead intel has even in their mobile ultra low voltage SKUs is just tremendous; while intel is pretty darn close in terms of efficiency. Probably too much to overcome in even several years time. If i'm wrong though, that's cool, it'll just drive the competition stakes up higher which will be fun to watch.

Exactly. However I see Apple worrying less about pure performance and more about something that is good enough. If they can get something with 70% performance but optimize it twice as well it should perform on par or better. Future is yet to be seen though so we don't know what'll happen. I am just predicting a merging of sorts at some point.

 

[saratoga172]

If people complain that Windows 8 is a bastardization of two UIs, why would folks consider an iOS/OSX hybrid to be any different/better?

Not saying they would or should. Just saying I see it heading that way. There are already elements of ios within OS X. And I think Apple would love to create one "merged" os and be able to use it across all or most devices. At least if I were a profit company I would.

Having said that when I look at a lot of Apple products I see them analyzing initial releases and taking and building onto it in such a way people like and find useful. The iphone is seen as revolutionary but we had touch devices before it. It simply took the device and slapped on a very intuitive and powerful OS and people bought into it. So for that I think they will look at Windows 8's shortcomings and the complaints and try to figure out how to avoid that. We may never see it. Time will tell.

 

[Enigmoid]

A7 hasn't leveraged its power yet because it's too power hungry to do so.

CPU only is using ~8W at 1.4 ghz clocks under a Prime 95 type load. NO GPU.

Watt for watt its slightly ahead of S800 and it looks like a little behind silvermont.

And really considering Apple's huge software and JS advantage its slower than you think.

If you look at anand's iphone 5 vs 5c benchmarks you see huge gains from IOS 7 (30-70%).

 

[Eug]

Is Merryfield actually even available yet?

Because if not, it may be competing against Apple's A8.

 

[Nothingness]

The ROB is just stupidly large.

Given the number of units and their pipeline depths, it's not that stupid.

 

[Rakehellion]

According to GeekBench, the i7-4770K gets a score of 15000, while the A7 is about 2500. If you could scale the A7 up to quad core and 3.4GHz it would be about on par.

 

[Nothingness]

According to GeekBench, the i7-4770K gets a score of 15000, while the A7 is about 2500. If you could scale the A7 up to quad core and 3.4GHz it would be about on par.

But the A7 probably can't scale to 3.4 GHz

 

[NTMBK]

The ROB is just stupidly large.

It's the same size as Haswell...

 

[NTMBK]

But the A7 probably can't scale to 3.4 GHz

But if they can scale it to ~2GHz, that would be enough to replace Intel in MacBooks.

 

[Homeles]

Given the number of units and their pipeline depths, it's not that stupid.

I wasn't intending people to take that literally.

It's the same size as Haswell...

Exactly.

 

[sefsefsefsef]

Exactly.

We know that the A7 is limited by its current form factors (iPad and iPhone), but we don't know exactly how much further it could go. As it stands, an i7 4770K is about twice as fast at SunSpider than an iPhone5S. Single threaded Geekbench3 Haswell performance at 3.9 GHz (Turbo) is 2.7x higher than an iPhone5S at 1.3 GHz (a 3.0x clockspeed difference).

If the A7 could be clocked a little higher in a different form factor (even just up to 2 GHz in a laptop), then IMO it would be a serious Haswell ULV competitor. I think it far outpaces any Atom or Kabini and, microarchitecturally at least, it's on the same level as Haswell.

 

[jdubs03]

But if they can scale it to ~2GHz, that would be enough to replace Intel in MacBooks.

They'll probably be about there with the A9, especially if they use TSMC's FinFet+ process. But, there really isn't anyway to tell. Even if the A8 has 50% higher performance (lets use GB 3 as the benchmark, which is beneficial to Apple as noted before IOS 7.0 helps a lot), the single-thread 64-bit score would be about 2100. I expect the lowest tier i5-u to be around 2700-2800, but when compared to the y-series the A8 could potentially be on par.

 

[witeken]

Looking at Cyclone makes one thing very clear: the rest of the players in the ultra mobile CPU space didn't aim high enough. I wonder what happens next round.

It's a very informative article, but I do not agree at all with this statement of him. Why didn't Qualcomm or ARM aim high enough when they have a number of architectures that are only able to run at under 1GHz when staying withing the TDP (= not using boost; sustained clock speed for an infinite amount of time).

Everyone is actually aiming way too high.

 

[witeken]

I think the likely next jump is can they and how do they integrate this into a tablet/computer hybrid. Will we see this running a version of OS X? For example.

The next jump is putting an Intel Core into the iPhone. If you're already using big cores, why not just take the absolute best? (cynicism)

Well, of course. At this time, yes, the i5 is better. But if apple can do i'm sure they would want to. It just not quite there yet.

His statement was more about the future, who knows what could happen there. Certainly no one is close to intel's core architecture at this point, but if it happens 5 years from now - hey - more competition and all that sort of thing. Intel is in a race to get their core architecture at even lower power points and they're doing amazingly well at that with their -Y mobile SKUs, while ARM SOC vendors are in a race to increase performance. Where this ends up will be interesting for sure, IMO. That said, I don't see mobile SOCs over-taking core anytime even remotely soon, not even in the next 5-7 years. The performance lead intel has even in their mobile ultra low voltage SKUs is just tremendous; while intel is pretty darn close in terms of efficiency. Probably too much to overcome in even several years time. If i'm wrong though, that's cool, it'll just drive the competition stakes up higher which will be fun to watch.

How do all those other companies - Apple, Qualcomm, Nvidia, ARM - want to overtake Intel when they have only a small fraction of the R&D budget, experience, and have to use increasingly more inferior process nodes?

 

[NTMBK]

The next jump is putting an Intel Core into the iPhone. If you're already using big cores, why not just take the absolute best? (cynicism)

Because they own the design entirely, giving them a competitive edge over their rivals with no access to it, and can pour their money into R&D instead of paying an external vendor to build it? An external vendor who will then sell the same chips to your rivals? Why fund their R&D when you can fund your own?

 

[DrMrLordX]

The next jump is putting an Intel Core into the iPhone. If you're already using big cores, why not just take the absolute best? (cynicism)

Topper's way ahead of you there.

 

[witeken]

Because they own the design entirely, giving them a competitive edge over their rivals with no access to it, and can pour their money into R&D instead of paying an external vendor to build it? An external vendor who will then sell the same chips to your rivals? Why fund their R&D when you can fund your own?

That's a fallacy. The wrong assumption you made is that Apple builds the best CPUs, but they don't. They're free to develop their own architecture, but why would they if they can't make the best? Silvermont has already proven to be vastly superior, so other companies can use it anyway while Apple sticks with their own inferior designs on inferior nodes.