A9 vs Intel

[Coolerooney]

Hi all

I was blown away with the A9 performance, checked some numbers, came up with this:

The A9 is Geekbenched at around 2500 single core @ 1,7 Ghz
The fastest Intel Core i7-4790K is measured 4470 @ 4 Ghz ( single core)

So for each Mhz the A9 is more efficient processor than the current Intel i7
If we could run the A9 @ 3,4 Ghz, the Geekbench would be around 5000- assuming the rest of the harddware can keep up

Off course the A9 cannot deal with this overclocking, however, considering that the Core i7 has coolers and fan, this might not be so far off. With the Bitcode tool crosslinking processors should be quite do-able.


Link to Apple Bitcode and the "Platforms State of the Union" video from the 2015 WWDC

https://medium.com/@InertialLemon/apple-s-bitcode-telegraphs-future-cpu-plans-a7b90d326228

Have fun!


Coolerooney




Please post your questions in the thread that's already there for A9 discussion.


esquared
Anandtech Forum Director

 

[asendra]

This has been covered ad nauseam in this forum. For example here http://forums.anandtech.com/showthread.php?t=2447200 (and if you read my comment there or in other similar posts you'll see that I believe apple SOCs are incredible, the A9 is a beast)

But a couple of quick things.

- You can't compare directly between arquitectures. Mobile SOCs have specialised hardware for certain things that skew the results in their favour, and intel processors also have specialised instructions that I believe are not being used in GB.
- A chip that does well at low frequency, it's arquitetured to do so, so it's not as easy as cracking up the MHz up to eleven.
- And that medium article is the biggest lie about bit code that everyone just keep repeating! This has been confirmed by non other than Chris Latter, creator of the LLVM compiler (and of swift). LLVM IRs are not completely cpu agnostic, so they don't allow to go from arm to x86 with a button switch.

If you want to know more about what bit code allows read this: https://news.ycombinator.com/item?id=9726552

I'll quote you Chris Lattner comment about bitcode:


drfuchs 99 days ago

I managed to ask Chris Lattner this very question at WWDC (during a moment when he wasn't surrounded by adoring crowds). "So, you're signaling a new CPU architecture?" But, "No; think more along the lines of 'adding a new multiply instruction'. By the time you're in Bitcode, you're already fairly architecture-specific" says he. My hopes for a return to big-endian are dashed. [Quotes are approximate.]

And I say all that as the biggest Apple user possible. I use a rMBP, an iPad, iPhone etc (never even touched Android apart from development).

 

[Coolerooney]

hi Asendra,

Yes I saw that thread prior to my posting.

Question: isn't Geekbench doing exactly that- comparing performance between different architectures?

Could it be that Chris Lattner is not telling the whole story?

Coolerooney

 

[asendra]

hi Asendra,

Yes I saw that thread prior to my posting.

Question: isn't Geekbench doing exactly that- comparing performance between different architectures?

The way I see it, Geekbench is best used comparing performance between devices of the same platform.
Certain subscores of the bench are better than others to compare platforms. Others can and will explain you better why.

Could it be that Chris Lattner is not telling the whole story?

Coolerooney

It's not only him, a lot more people with experience in LLVM, and IRs say the same thing.

 

[Coolerooney]

I agree with you that it may be difficult to compare Apples to apples...

Has anybody hooked up a Ipad air 2 /iphone 6 to a VGA adapter/ monitor/ iOS mouse?

If yes what were the outcomes?

 

[Phynaz]

hi Asendra,

Yes I saw that thread prior to my posting.

Then what was the point of creating another thread to discuss the exact same thing?

 

[Coolerooney]

..

 

[Eug]

I agree with you that it may be difficult to compare Apples to apples...

Has anybody hooked up a Ipad air 2 /iphone 6 to a VGA adapter/ monitor/ iOS mouse?

If yes what were the outcomes?

No, but using an iPad Air 2 (2 GB RAM and iOS 9.1) with a Bluetooth folding keyboard works well, aside from the fact I have no mouse. The Air 2 often feels more responsive than a 2.0 GHz Core Duo iMac (2 GB RAM and Mac OS X 10.6) but I dislike using the touchscreen for navigation when using a keyboard with a tablet.

Similarly, this isn't exactly an Apples to apples comparison, but my Android phone from 2012 with dual-core ARM Krait 1.5 GHz and 1 GB RAM hooked up to a 720p monitor running Jelly Bean felt much more responsive for surfing (without Flash) than my Atom 330 dual-core 1.6 GHz machine with ION and 3.12 GB RAM running Windows 7 for surfing (with or without Flash). The main problem with the Android phone setup is that the OS is not built for non-tablet use, so navigating around the screen is a pain.

---

tl;dr:

Current ARM solutions running tablet OSes feel very responsive for basic usage like surfing and email. The main issue isn't the CPU performance. The main issue is the fact that the OSes aren't built for laptop or desktop type use because they assume you'd be using the touchscreen.

---

That said, there are situations where higher CPU performance in an ARM tablet is welcome. However, these are in the context of (mild) content creation. For example, it'd be nice if iMovie video exports were faster. Such video encoding is actually important in a phone, since current ARM phones actually record 4K video.