Next-Gen iPhone & iPhone Nano Speculation

[anexanhume]

2*A9@1.6GHz probably doesn't make that much sense under the light that other SoCs at the moment don't have as huge GPU blocks amongst other aspects. Other than that the OMAP4470 has 2*A9@1.5GHz under 45nm (and a much more humble single SGX544@384MHz). I assume the tricky part is the GPU block and its respective power consumption; however the CPU block itself singled out is anything but absurd for 2*A9@1.6GHz under 32nm.

For the record ST Ericsson has two dual A9 mainstream SoCs on its roadmap both under 28nm if memory serves well. One clockes the 2*A9 at up to 1.85GHz and the other up to 2.3GHz and yes 28nm is a totally different chapter.

Yes it's unusual for Apple and you're right in the notion that they rarely invest in frequency but there's also a lot of pressure from the competition. Theoretically 2*A9@1.6GHz doesn't sound like an as much wild scenario; personally I'd prefer of course the 2*A15 scenario at more humble frequencies but that's besides any point.

Don't forget the posted app specific improvements: http://d35lb3dl296zwu.cloudfront.net/uploads/photo/image/8181/iPhone5_0254.jpg

So if it opens the pages app 2.1X faster, you'd need 1.68 GHz core clock assuming no bottleneck/speed-up from the RAM. It's possible they increased core clocks and tweaked the RAM type and memory interconnects, but that seems like a lot of effort for a current architecture when I don't feel they act as though they're under pressure from competitor's SoCs.

Also, regarding the 544, isn't that two 543's lashed together with some directX support thrown on top?

 

[ltcommander.data]

True. However it's still not good enough indication for a preliminary A15 conclusion. It's not like you can't in theory reach twice the number crunching with a high clocked dual A9. It could be truly be that someone reliable told Anand at the event; that's of course a totally different story. Based just on the presentation (from what I've seen so far) and the A6 moniker I don't see anything that guarantees a dual A15 unless I've missed something.

1.6GHz A5-based CPU would be "2x faster"

I was mainly referring to comments that improved flash memory performance may contribute to the 2x figure. While better flash I/O may be a factor in the 2x real-world performance examples that Apple cites, I think the 2x faster CPU claim refers to the CPU itself. How the 2x CPU increase is achieved via clock speed or architecture change is still up in the air, I just don't think they are roping flash I/O into their CPU claims. I could certainly see them improving the flash I/O to ensure that the 2x theoretical CPU performance increase actually translates into a 2x real world performance increase in when using the device/apps.

Also, regarding the 544, isn't that two 543's lashed together with some directX support thrown on top?

I believe a single SGX544 is a single SGX543 with DX9 support and not two SGX543.

 

[tangey]

I believe a single SGX544 is a single SGX543 with DX9 support and not two SGX543.

Indeed

 

[Pressure]

Hope anandtech isn't your source for that, as he states:-
"generally referred to as A5R2"
Which sounds a million miles from an official apple name

http://www.anandtech.com/print/6281

Would you feel better if I called it S5L8942 then?

 

[french toast]

I think they would obviously go for the option of the least power consumption considering the form factor...that might be 2*a9s @1.6ghz....or it just as well may be 2*a9s @ 1.2ghz with better ram,fsb,IO etc...which would easily load things up 2x better especially if ios 6 includes better memory management/speed to go along side the expected increase in ram...

I would think that would be a viable option for them if it increased efficiency.

 

[Gubbi]

Don't forget the posted app specific improvements: http://d35lb3dl296zwu.cloudfront.net/uploads/photo/image/8181/iPhone5_0254.jpg

So if it opens the pages app 2.1X faster, you'd need 1.68 GHz core clock assuming no bottleneck/speed-up from the RAM. It's possible they increased core clocks and tweaked the RAM type and memory interconnects, but that seems like a lot of effort for a current architecture when I don't feel they act as though they're under pressure from competitor's SoCs.

Also, regarding the 544, isn't that two 543's lashed together with some directX support thrown on top?

So the improvement is starting of apps (loading them) and storing images. Smells like slightly faster CPU with more bandwidth (both to RAM and FLASH)

I'm guessing A6 is a die shrunk A5X + some accelerators.

Let's see some number crunching benchmarks, heck, even Sunspider will do.

Cheers

 

[anexanhume]

Indeed

My mistake, I was thinking the 554 was the 544.

So the improvement is starting of apps (loading them) and storing images. Smells like slightly faster CPU with more bandwidth (both to RAM and FLASH)

I'm guessing A6 is a die shrunk A5X + some accelerators.

Let's see some number crunching benchmarks, heck, even Sunspider will do.

Cheers

We don't even need that. As soon as someone has one in hand they just need to run the app that reports CPU speed and RAM size. CPU clock should be able to tell us.

 

[Arun]

I will say one thing, and one thing only.

If AnandTech tries using a non-NEON Linpack (or even worse Java-based Linpack on Android) to determine whether it's an A9 or A15 based on NEON performance (which Linpack doesn't use) then I will be very depressed.

 

[anexanhume]

I will say one thing, and one thing only.

If AnandTech tries using the Java-based Linpack to determine whether it's an A9 or A15 based on NEON performance (which Linpack doesn't use) then I will be very depressed.

I have more faith in Anand than that.

But you'll tell us the GPU, right?

Actually, looking at GPU, it has to be frequency doubled 543MP2 or die shrunk MP4. Minimum Rogue performance should be 5X from MP2 going by triangle count alone.

 

[Arun]

But you'll tell us the GPU, right?

I can confidently say it renders triangles, and does so far faster than I can do so by hand with crayons. And please don't ask about my crayon drawing speed because that might get me in trouble

 

[MarkoIt]

Some more founded guess:
http://arstechnica.com/apple/2012/09/what-do-we-know-about-apples-new-a6-processor/

 

[silent_guy]

Not that I wouldn't like a faster CPU than the one in my trusty, soon to be retired, iPhone 4, but isn't app startup time one of the most important parameters? Ever since the introduction of the 3GS, everything has already been pretty smooth. Startup times are one of the things that still annoy me at times.

So if that's been reduced by half: give it to me now.

I'm trying to see what the next gen will have? I can't think of anything compelling that the current generations of phones (in general) doesn't have.

Wireless charging? Meh: I prefer this nice looking aluminum brushed dock on my nightstand over some ugly mat. What else? We're already maxed out on possible sensors. LTE is way faster than my more than sufficient home Internet connection. CPU speed has been fine already for the last generation. GPU too. I'll have the new screen size feels, not really excited about it. Better camera maybe? What else?

 

[Exophase]

But you'll tell us the GPU, right?

Since he already made that same error with Krait I wouldn't give him that much confidence..

 

[anexanhume]

I can confidently say it renders triangles, and does so far faster than I can do so by hand with crayons. And please don't ask about my crayon drawing speed because that might get me in trouble

I don't need to know how fast you can draw. That's a process that varies. I just need to know how many hands you draw with.

Some more founded guess:
http://arstechnica.com/apple/2012/09/what-do-we-know-about-apples-new-a6-processor/

Their sizing analysis is good to suggest A15, but their GPU claim is questionable. G6200 is 5X the triangle performance of 543MP2.

 

[Jubei]

Some more founded guess:
http://arstechnica.com/apple/2012/09/what-do-we-know-about-apples-new-a6-processor/

Interesting. They exclude quad-cpu but they never entertain the thought of dual A9s clocked high.

A6 = 22% smaller than 45nm A5

32nm A5 = 41% smaller than 45nm A5

32nm A5X = theoretically 21% smaller than 45nm A5

Based on this we can exclude a simple overclocked 32nm A5 no? Because either GPU or the CPU has to be bigger in A6 to be only 22% smaller

 

[Exophase]

Some more founded guess:
http://arstechnica.com/apple/2012/09/what-do-we-know-about-apples-new-a6-processor/

A6 being really close to their projected 32nm A5X size seems to be a good argument that it's a 32nm A5X, with a lot more CPU clock speed. I don't think they realize that Cortex-A15s are probably substantially bigger than Cortex-A9s.

This all reminds me of when A4 was first announced and a lot of people were saying it has Cortex-A9 inside.

 

[Ailuros]

I don't need to know how fast you can draw. That's a process that varies. I just need to know how many hands you draw with.

Judging by that monstrosity: http://www.beyond3d.com/content/reviews/1/14 he's using at least as many hands as clusters you can scale on Rogue

I'm willing to bet that Rys has hidden from him all the crayons ever since.

 

[silent_guy]

Is there a fundamental advantage of having a low clocked A15 vs. a high clocked A9? That is: as a customer, and not a tech geek, should I care?

 

[Exophase]

Is there a fundamental advantage of having a low clocked A15 vs. a high clocked A9? That is: as a customer, and not a tech geek, should I care?

You'd have to look at perf/W curves to really get an idea. Since these curves are not linear even an intrinsically much more power hungry CPU might use less power at a much lower clock if it can match performance there. This is more true as you pass the ideal knee for some processor.

Of course the performance differences are hardly uniform. If Apple cares a lot about SIMD performance A15 should be a bigger win than it would be otherwise.

 

[zed]

Im surprised that apple shares went up 3%. In all honesty this phone unveil is prolly less exciting unveil than the 4s (which wasnt that great, though it did have siri which made the masses go ohhh). I wonder how it will sell? samsung s3 was 20million in ~3 months, if it doesnt ~double that we may be in the process of the crown being passed.