Apple A9 SoC

[Lazy8s]

In one of the presentations during the Keynote (maybe the Adobe one, I seem to remember it being an external dev versus one of the Apple execs), the presenter clearly mentioned one of the benefits of the iPad Pro beings its increased memory (as opposed to the memory bandwidth and flash access upgrades mentioned by Schiller separately). That had me expecting 4 GB.

It may be futile, though, trying to re-watch for the segment in the presentation for confirmation since Apple apparently has a team standing by right after the live show to get the presenters to re-record any mistakes/slips they made and then replace it in the archive. I discovered this myself last year in the Pixelmator (I think that's what it was called) iPad Air 2 presentation. This year, Schiller slipped up once when he used the term Force Touch instead of 3D Touch at one point. That's probably already been edited in the archived video.

I don't take the possibility of anything other than 2 GB for the new Apple TV and iPhones seriously this year. I felt like that about the new iPhones last year, though... so anything's possible I guess. It would be more counterproductive to the evolution of their platforms this year, though, to stick to 1 GB, so I seriously doubt it. A TV platform with some level of focus on games needs more RAM, and the new iPhones with less RAM would splinter the dev targets too much.

The Watch, iPad Pro, and new Apple TV seem to miss the mark in my eyes as products. I'm not expecting big things from any of them. I feel Apple did a great job with the hardware of the iPad Pro: its display, form factor, A9X CPU and GPU, memory specs, and peripherals all turned out impressively. Still, it won't sell to the growing number of people who require a personal computer for specific, established productivity purposes and want a device that makes for both a great tablet and laptop. It's left to be an oversized iPad, which hadn't been selling to their expectations already.

My guess for the A9X is 667 MHz GT7800 + 2 GHz tri-core Cyclone v3. For that to work instead of a lower clocked hypothetical 12 cluster "GT7850" combination, the benefit of the memory bandwidth and promised architectural upgrades from Series 7 and on the CPU side would've really needed to deliver.

My guess for the A9 is a 500 MHz GT7600 + 1.5 GHz tri-core Cyclone v3.

3D Touch (terrible name to use especially when they already established Force Touch) will be great for enhancing the already solid gesture based navigation. I especially like how I could swipe in from the left side of the screen to go back a page in an app or simply push a little harder now and go back between apps themselves, creating an intuitive extension to what they've already built. I'm overall fairly impressed with the new iPhones and plan to pick up the 6s Plus for the new processor platform. Apple didn't seem to highlight much in the way of 6s Plus exclusive features versus the regular 6s this year, so I wonder if a marginally higher GPU clock will be there. Maybe just the higher throttling point this year.

 

[Pressure]

In one of the presentations during the Keynote (maybe the Adobe one, I seem to remember it being an external dev versus one of the Apple execs), the presenter clearly mentioned one of the benefits of the iPad Pro beings its increased memory (as opposed to the memory bandwidth and flash access upgrades mentioned by Schiller separately). That had me expecting 4 GB.

It may be futile, though, trying to re-watch for the segment in the presentation for confirmation since Apple apparently has a team standing by right after the live show to get the presenters to re-record any mistakes/slips they made and then replace it in the archive. I discovered this myself last year in the Pixelmator (I think that's what it was called) iPad Air 2 presentation. This year, Schiller slipped up once when he used the term Force Touch instead of 3D Touch at one point. That's probably already been edited in the archived video.

I don't take the possibility of anything other than 2 GB for the new Apple TV and iPhones seriously this year. I felt like that about the new iPhones last year, though... so anything's possible I guess. It would be more counterproductive to the evolution of their platforms this year, though, to stick to 1 GB, so I seriously doubt it. A TV platform with some level of focus on games needs more RAM, and the new iPhones with less RAM would splinter the dev targets too much.

The Watch, iPad Pro, and new Apple TV seem to miss the mark in my eyes as products. I'm not expecting big things from any of them. I feel Apple did a great job with the hardware of the iPad Pro: its display, form factor, A9X CPU and GPU, memory specs, and peripherals all turned out impressively. Still, it won't sell to the growing number of people who require a personal computer for specific, established productivity purposes and want a device that makes for both a great tablet and laptop. It's left to be an oversized iPad, which hadn't been selling to their expectations already.

My guess for the A9X is 667 MHz GT7800 + 2 GHz tri-core Cyclone v3. For that to work instead of a lower clocked hypothetical 12 cluster "GT7850" combination, the benefit of the memory bandwidth and promised architectural upgrades from Series 7 and on the CPU side would've really needed to deliver.

My guess for the A9 is a 500 MHz GT7600 + 1.5 GHz tri-core Cyclone v3.

3D Touch (terrible name to use especially when they already established Force Touch) will be great for enhancing the already solid gesture based navigation. I especially like how I could swipe in from the left side of the screen to go back a page in an app or simply push a little harder now and go back between apps themselves, creating an intuitive extension to what they've already built. I'm overall fairly impressed with the new iPhones and plan to pick up the 6s Plus for the new processor platform. Apple didn't seem to highlight much in the way of 6s Plus exclusive features versus the regular 6s this year, so I wonder if a marginally higher GPU clock will be there. Maybe just the higher throttling point this year.

Well, the App Programming Guide for tvOS clearly says 2 GB, so it is confirmed.

• 64-bit A8 processor
  
  
• 32GB or 64GB of storage
  
  
• 2GB of RAM
  
  
• 10/100Mbps Ethernet
  
  
• WiFi 802.11a/b/g/n/ac
  
  
• 1080p resolution
  
  
• HDMI
  
  
• New Siri Remote / Apple TV Remote

 

[anexanhume]

My guess for the A9 is a 500 MHz GT7600 + 1.5 GHz tri-core Cyclone v3.

The die shot looks like it shows two CPU cores. They mention in the keynote they spent a lot of time trying to make your everday tasks faster, so maybe they wont on a profiling binge on their OS and some common applications. With control over the language, compiler, OS and uarch, I'd say they are pretty well positioned to get some nice optimization speedups.

 

[tangey]

Slightly OT, but I notice the most recent entry in the Khronos gles conformance table, is from Apple. It's a single listing that applies to iphone6,iphone6+, itouch 6 and...wierdly ipad air 2. It lists the processor as A8, and the GPU as A8, driver 53.13
https://www.khronos.org/conformance/adopters/conformant-products/#opengles

The previous entry for the ipad air2, which was a separate entry on it's own, (scroll down the above link to the 21st of Nov 2014) listed the processor correctly as the A8X.

I wonder what's going on there, I guess someone made a mistake in the submission to Khronos.

 

[Megadrive1988]

For iPad Pro, it'll be be really interesting to see which PowerVR7XT config. they're using in A9X.

 

[iMacmatician]

I think Entropy was refering to the fact that a core count increase would represent a quite a bit larger step for the A9 vs. A9X. So, if the A9 had 3 cores instead of 2, that would already be a 50% improvement, leaving only ~5% for IPC improvements + ~7% for higher clocks or so to get to the 70% vs A8. But for the A9X, a core count increase to 4 cores would only represent a 33% improvement, so with the same 5% IPC improvements it would need ~30% higher clock to get the 80% against A9.
Who knows though what metric those 70%/80% increases refer too. It is also possible the chips have quite a bit higher peak clock but need to throttle more than they had to in the past, so using the right workload just about any quoted percentage increase would sound reasonable for both of them, with or without more cores for either of them...

Last year, the CPU performance increase from the A7 (iPad) to the A8X was advertised as 40% despite the extra core in the latter*. Given that the increase from the A7 (iPhone) to the A8 was 25% (and similar clock speed jumps in both situations), that leaves only 10-15% for the third core. I think the +40% was from a workload or combination of workloads that mainly used 1 or 2 threads. So I don't expect core count increases to factor a lot into these numbers unless Apple is measuring the performance increases differently this year or something else is going on.

* Apple also mentioned a 2x increase in certain situations, and while I don't think a similar statement was made yesterday, I'm not taking its absence as necessarily an indicator of anything.

 

[Entropy]

Last year, the CPU performance increase from the A7 (iPad) to the A8X was advertised as 40% despite the extra core in the latter*. Given that the increase from the A7 (iPhone) to the A8 was 25% (and similar clock speed jumps in both situations), that leaves only 10-15% for the third core. I think the +40% was from a workload or combination of workloads that mainly used 1 or 2 threads. So I don't expect core count increases to factor a lot into these numbers unless Apple is measuring the performance increases differently this year or something else is going on.

While this is true, the provisions for user level multitasking in the new iOS may (or may not) have caused them to change their performance estimation practises. We are engaging in a game of plausible speculation. Enjoy the entertainment and intellectual stimulation while it lasts. So if Apple profiling causes them to attach a modest value to adding cores, then the task remains to try to figure out the combination of IPC and frequency improvements that can make up for an 80% increase in CPU performance for the A9x for instance. What changes could they have done this year to boost IPC?

 

[Squilliam]

Confirmation bias...

These things aren't "everywhere"; the Surfaces are a niche product, which is why MS is throwing their weight behind iOS app development. If MS's hardware sales were awesome, they wouldn't do that. It's not their standard modus operandi to support competitors' ecosystems unless they have to. (It isn't Apple's either one might observe.)

At university I have noticed that they form the second highest 'species' of laptops behind Macs. I think Microsoft is making some headway there.

 

[Ailuros]

[

For iPad Pro, it'll be be really interesting to see which PowerVR7XT config. they're using in A9X.

Mirror mirror on the wall..... if the die shot Nebu refers to should be for real and it's a 7600 in the A9, it smells suspiciously like a 2*GT7600 config in the A9X.

 

[rapso]

the 80% can also be a result of doubling the memory performance. I think the A7 was quad channel 32bit DDR2, then they went to single 64bit DDR3, A8x was dual 64bit DDR3, A9(x) could be quad 64bit DDR3. I don't think I've ever seen apple telling explicitly at a show that they've doubled memory performance. twice bandwidth -> twice speed CPU and GPU (I'm sure everyone can write a benchmark that proofs that to be correct ).

 

[tangey]

[


Mirror mirror on the wall..... if the die shot Nebu refers to should be for real and it's a 7600 in the A9, it smells suspiciously like a 2*GT7600 config in the A9X.

Would the same reasoning be in place this time as last time ? In series 6, they had nothing public that could do 2*6450. IMG have public 7XT IP that can hit the performance of 2*GT7600. An upclocked 7800 could easily do it, and given the much bigger chassis, upclocking would be very possible. I guess it depends on how much die-space they want to use, in that chip

Going off-topic, did Apple actually state a typical battery life for the ipad pro? , I can't remeber seeing it.

 

[rapso]

did Apple actually state a typical battery life for the ipad pro?

I think they've stated 10h (a full day)


http://www.apple.com/ipad-pro/specs/

All models

• Built-in 38.5-watt-hour rechargeable lithium-polymer battery
• Up to 10 hours of surfing the web on Wi‑Fi, watching video, or listening to music
• Charging via power adapter or USB to computer system

Wi-Fi + Cellular model

• Up to 9 hours of surfing the web using cellular data network

 

[Laurent06]

the 80% can also be a result of doubling the memory performance. I think the A7 was quad channel 32bit DDR2, then they went to single 64bit DDR3, A8x was dual 64bit DDR3, A9(x) could be quad 64bit DDR3. I don't think I've ever seen apple telling explicitly at a show that they've doubled memory performance. twice bandwidth -> twice speed CPU and GPU (I'm sure everyone can write a benchmark that proofs that to be correct ).

I bet on LPDDR4 (this is Apple, we can only guess and bet ).

 

[rapso]

I bet on LPDDR4 (this is Apple, we can only guess and bet ).

would also explain why they were talking about desktop performance, afaik skylake with DDR4 won't have 50GB/s.

quad chan LPDDR4 in A10(x) -> 100GB/s

 

[french toast]

Rumours coming out of china point to a 2 plus 2 cpu design, not unlike big little. One low power cluster running at 1.2ghz and the high powered cluster running at 1.7ghz, with 2 gig ram.
Couple of thoughts, die would be massive If using cyclone for all cores would it not? Unless using a53s in traditional ARM style.._so un apple like.
Wffctech

 

[Laurent06]

Rumours coming out of china point to a 2 plus 2 cpu design, not unlike big little. One low power cluster running at 1.2ghz and the high powered cluster running at 1.7ghz, with 2 gig ram.
Couple of thoughts, die would be massive If using cyclone for all cores would it not? Unless using a53s in traditional ARM style.._so un apple like.
Wffctech

The Cyclone cores, even if rather big, are small when looking at all of the SoC: http://www.anandtech.com/show/8562/chipworks-a8

 

[french toast]

The Cyclone cores, even if rather big, are small when looking at all of the SoC: http://www.anandtech.com/show/8562/chipworks-a8

Yea thanks thought they took up a larger % than that, still if you factor in the likely extra gpu clusters, new ( larger?) memory controller , cache increase? And the fact transistor density cant have improved that much between tsmc 20nm and samsung 14nm LPE, it does seem a stretch IMO.
From an efficiency perspective however it would be brilliant.

 

[mczak]

the 80% can also be a result of doubling the memory performance. I think the A7 was quad channel 32bit DDR2, then they went to single 64bit DDR3, A8x was dual 64bit DDR3, A9(x) could be quad 64bit DDR3. I don't think I've ever seen apple telling explicitly at a show that they've doubled memory performance. twice bandwidth -> twice speed CPU and GPU (I'm sure everyone can write a benchmark that proofs that to be correct ).

Apple did say they doubled memory bandwidth, and we know it's 128bit (really 4x32bit) LPDDR4.
This, however, will do nothing for cpu performance. Unless you measure memcpy (and even then I'm actually sceptical the cpu cores can actually saturate all the bandwidth).

 

[Lazy8s]

The newer drivers on iOS 9.1 Public Beta (75.10.6) improve ES 3.0 performance in GFXBench by a few percent on last year's devices, so there's a little there.

 

[ltcommander.data]

the 80% can also be a result of doubling the memory performance. I think the A7 was quad channel 32bit DDR2, then they went to single 64bit DDR3, A8x was dual 64bit DDR3, A9(x) could be quad 64bit DDR3. I don't think I've ever seen apple telling explicitly at a show that they've doubled memory performance. twice bandwidth -> twice speed CPU and GPU (I'm sure everyone can write a benchmark that proofs that to be correct ).

The A7 was 2 x 32-bit LPDDR3, the A8 was 2 x 32-bit LPDDR3, and the A8X was 4 x 32-bit LPDDR3. Apple hasn't used a single channel memory controller since the A4.

Rumours coming out of china point to a 2 plus 2 cpu design, not unlike big little. One low power cluster running at 1.2ghz and the high powered cluster running at 1.7ghz, with 2 gig ram.
Couple of thoughts, die would be massive If using cyclone for all cores would it not? Unless using a53s in traditional ARM style.._so un apple like.
Wffctech

The Apple Watch uses a customized Cortex A7 core running armv7k. Apple could build a customized A53 core that can serve on its own in a future Apple Watch or be used as a companion core in iPhones/iPads so they can have a core meeting their needs without splitting their resources on too many designs.