Do we yet have confirmed information about A10 memory bandwidth? 51.2 GB/s like A9X or are we back at 25.6 GB/s? None of the reviews mentioned this.
 
Do we yet have confirmed information about A10 memory bandwidth? 51.2 GB/s like A9X or are we back at 25.6 GB/s? None of the reviews mentioned this.

Chipworks confirms it is like 6s.

"The A10 sits below the Samsung K3RG1G10CM 2-GB LPDDR4 memory. This is similar to the low power mobile DRAM as the one we found in the iPhone 6s."
 
Chipworks confirms it is like 6s.

"The A10 sits below the Samsung K3RG1G10CM 2-GB LPDDR4 memory. This is similar to the low power mobile DRAM as the one we found in the iPhone 6s."
Both iPhone 7 and 7+? iPhone 7 has 2 GB of memory while 7+ has 3 GB memory... 38.4 GB/s could be theoretically at least possible in 7+ (triple channel).
 
Could "Hurricane" be the codename for the A10's CPU (I assume the big core)?

TechTastic points out that Apple's list of kernel changes in macOS 10.12 contains a reference to "Hurricane":
macOS 10.12 API Diffs said:
Added #def CPUFAMILY_ARM_HURRICANE
While TechTastic speculates about the possibility of ARM Macs, the OS X 10.11 kernel changes also contains
OS X v10.11 API Diffs said:
Added #def CPUFAMILY_ARM_TYPHOON
so I'm not sure as to what extent this info points to a possible ARM transition.
 
It's pretty obvious Apple has ported MacOS X years ago, but that doesn't mean a machine with it is coming.

True but I have been wondering why they deviated from their historical naming convention and named it A10 Fusion. Maybe hinting that it could be used in other devices or am I just reading too much into it?

The performance figures so far are certainly promising and with a higher TDP should compete well with Intel's Y series processors. The low power cores also allowed them to design the bigger cores for higher speed and this would be useful in a larger form factor. A candidate for at least the 12" Macbook maybe?
 
Last edited:
From Barron's: "Apple’s ‘A10′ iPhone Chip Smokes the Competition, Says Linley Group."

30462581965_ddd5072329.jpg


Zephyr is the name of the little core in the A10.
 
So A10 big but still 3 times smaller as an Intel core.

I saw a comment about this elsewhere about that area including an L3 slice. I did some rough area calculations on an Intel die shot that seems to match this claim, as core+L2 is around 7mm2.
Intel's core could be argued to need at least some L3, but I think the same could be said for Apple's L2. Its core is still smaller, but not to the same extent.
 
From PC World: "The mysteries of the GPU in Apple's iPhone 7 are unlocked."

PC World said:
The GPU in the iPhone 7 uses a custom version of the PowerVR GT7600 GPU, which is based on the same graphics processor architecture as in last year’s iPhone 6S and 6S Plus, according to an analysis by The Linley Group, which specializes in semiconductors.
[…]
“We believe the iPhone 7, to avoid overheating, throttles back from its top GPU speed after a minute or less, preventing it from achieving a high score for all users,” Gwennap [founder and principal analyst at Linley Group] said.

The A10 Fusion GPU has six cores and operates at around 900MHz. It cranks up the frequency to reach peak speed.
 
The A10 has now been in the wild for several months.

Is there any signficance to the fact that this chip (unlike all previous iphone Socs) has not been submitted to Khronos for OpenGles conformancy testing ?
 
The A10 has now been in the wild for several months.

Is there any signficance to the fact that this chip (unlike all previous iphone Socs) has not been submitted to Khronos for OpenGles conformancy testing ?

I note that in Jon Peddie's wide ranging article on the whole IMG/Apple thing, he says that Apple:-
As is well known, Apple used the concepts of the classic binding model from OpenGL and applied it to OpenCL. In so doing, Apple abandoned the Khronos industry standard OpenGL and OpenGL ES as well as Vulkan API, all of which point to that vertical integration thing

Clearly they are still supporting all the right calls, as otherwise the benchmarks would fall over, but it does explain why they are no longer submitting to khronos.

That article can be seen here:-
http://www.eetimes.com/document.asp?doc_id=1332007
 
Last edited:
Suppose that the big core in the A11 (for example) features simultaneous multithreading. Then (assuming the little cores don’t have SMT) does it make more sense for each big core to be paired with…
  1. one little core, or
  2. as many little cores as there are threads in the big core?
 
SMT isn't going to be used in mobile any time soon.

All Atom phone/tablet SoCs up until Cloverview had hyperthreaded cores, so technically it's being used right now (e.g. for people using any first-gen Zenfone).

Also, I don't know how broad the term "mobile" can be to you, but AMD's Banded Kestrel has a starting TDP of 4W and it uses 2 cores with hyperthreading. Should Microsoft ever release a phone-able Windows 10 version for x86 (no need to emulate win32 code), I could see some daring OEMs like Asus launching a mobile phone with a Banded Kestrel.
 
All Atom phone/tablet SoCs up until Cloverview had hyperthreaded cores, so technically it's being used right now (e.g. for people using any first-gen Zenfone).

Also, I don't know how broad the term "mobile" can be to you, but AMD's Banded Kestrel has a starting TDP of 4W and it uses 2 cores with hyperthreading. Should Microsoft ever release a phone-able Windows 10 version for x86 (no need to emulate win32 code), I could see some daring OEMs like Asus launching a mobile phone with a Banded Kestrel.
Mobile is about efficiency and HT makes no sense in that regard. The power penalty outweighs the performance benefit. ARM was generally clear that they see no point of SMT in mobile.
 
Back
Top