Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.

chrmjenkins

macrumors 603
Original poster
Oct 29, 2007
5,325
158
MD
Happy apple day everyone. Time for one final prediction update going into the event. In my previous prediction, I took none of the PCB leaks into account for my guess. Now we've received visual information from the package that shows it's smaller size than the A7 as well as a silk screen code that shows LPDDR3 at 1600 MHz in size is still in use. Somewhat surprisingly, it's also only 1GB still. The codes don't give any clear indication as to whether the fab house is now TSMC. The smaller pinout suggests many possibilities, as was discussed in the original post. Keep in mind that package size and pinout don't maintain a fixed ratio to die size, however.

Last night, a geek bench test bench run on a purported iPhone 6 showed a 1.38 GHz clock frequency, a modest bump over the 1.3 GHz A7 as expected, along with a roughly 10% speed boost, clock for clock. This was expected based on the ambitiousness of the A7 design already. It also seems to have confirmed 1GB RAM again.

Without further ado, the new prediction:

A8 Prediction
A8
  • Manufacturer - TSMC on HKMG 20nm process
  • Die Size - 90-100 mm2
  • Designer - Apple
  • CPU Type - 1.4GHz Third Generation Custom Apple Core
  • Core Count - 2
  • Instruction Set - ARMv8-A (with custom Apple extensions)
  • Chip Designator - S5L8970X
  • L1 Cache - 64/64KB
  • L2 Cache - 1MB
  • L3 Cache - 4MB
  • RAM - 1GB LPDDR3 @ 1600 MHz (64 bit interface, PoP for iPhone)
  • Max Theoretical Memory Bandwidth - 12.8 GB/s
  • GPU Type - "Hex Cluster" PowerVR GX6650 @ 450 MHz
  • GPU Performance - 172.8/354.6 GFlops (FP32/FP16)
 

note235

macrumors 6502a
Jun 29, 2007
585
20
Happy apple day everyone. Time for one final prediction update going into the event. In my previous prediction, I took none of the PCB leaks into account for my guess. Now we've received visual information from the package that shows it's smaller size than the A7 as well as a silk screen code that shows LPDDR3 at 1600 MHz in size is still in use. Somewhat surprisingly, it's also only 1GB still. The codes don't give any clear indication as to whether the fab house is now TSMC. The smaller pinout suggests many possibilities, as was discussed in the original post. Keep in mind that package size and pinout don't maintain a fixed ratio to die size, however.

Last night, a geek bench test bench run on a purported iPhone 6 showed a 1.38 GHz clock frequency, a modest bump over the 1.3 GHz A7 as expected, along with a roughly 10% speed boost, clock for clock. This was expected based on the ambitiousness of the A7 design already. It also seems to have confirmed 1GB RAM again.

Without further ado, the new prediction:

A8 Prediction
A8
  • Manufacturer - TSMC on HKMG 20nm process
  • Die Size - 90-100 mm2
  • Designer - Apple
  • CPU Type - 1.4GHz Third Generation Custom Apple Core
  • Core Count - 2
  • Instruction Set - ARMv8-A (with custom Apple extensions)
  • Chip Designator - S5L8970X
  • L1 Cache - 64/64KB
  • L2 Cache - 1MB
  • L3 Cache - 4MB
  • RAM - 1GB LPDDR3 @ 1600 MHz (64 bit interface, PoP for iPhone)
  • Max Theoretical Memory Bandwidth - 12.8 GB/s
  • GPU Type - "Hex Cluster" PowerVR GX6650 @ 450 MHz
  • GPU Performance - 172.8/354.6 GFlops (FP32/FP16)
Thanks!

This announcement is too exciting. Couldn't sleep. But the geekbench was a disappointment.
 

iMacmatician

macrumors 601
Jul 20, 2008
4,249
55
GPU Type - "Hex Cluster" PowerVR GX6650 @ 450 MHz
AnandTech thinks the GPU is likely to be a GX6650. Given Apple's performance claims the GPU is probably around that clock speed too.

Good work on the predictions! So far they seem consistent with what's been unveiled.
 

chrmjenkins

macrumors 603
Original poster
Oct 29, 2007
5,325
158
MD
AnandTech thinks the GPU is likely to be a GX6650. Given Apple's performance claims the GPU is probably around that clock speed too.

Good work on the predictions! So far they seem consistent with what's been unveiled.

We need to be careful, because the G6630 has exactly 50% more processing power than the current G6430, but ImgTec claims the xx50 architectures are 50% faster. So, it depends on whether Apple hinges on that claim or not. Given the xx50 actually do introduce more FP16 performance, and it's not just a menagerie of ambiguous architecture improvements, I'm guessing both didn't' happen. Given the huge jump in transistors yet again to 2b, it seems likely they spent more transistors on the GPU, giving us the G6630. Now we just have to see if the A8 for iPads is the exact same.

Some other odds and ends:
FCC reports for Sprint version: https://apps.fcc.gov/oetcf/eas/repo...ame=N&application_id=664225&fcc_id=BCG-E2817A

Anandtech article covering the SoC announcement, they give a die size estimate of 89mm^2 going off of the 25% smaller number Apple quoted (not sure how their math works there): http://www.anandtech.com/show/8504/apple-announces-a8-soc

The jump to 2b transistors is quite impressive given they managed to shrink the die too, given TSMC claims a transistor density of 1.9x from 28nm to 20nm (assuming Samsung 28nm = TSMC 28nm of course). It seems likely they increased that SRAM L3 cache to 8MB, adding another ~250m transistors by itself. They could have also added some functionality that was previously off-chip, as I suggested. We'll know when we find or don't find the M8, for example.

Their speed improvements suggest modest architecture improvements, especially if that 1.38 GHz clock speed from geek bench is accurate.

It also makes sense that power went down and they claim better efficiency. A smaller package means fewer pins for power, so you would expect it to go down. It's interesting though that the LTE browsing is no better on the iPhone 6 than the iPhone 5S, despite all other numbers being better. They must be assuming full bandwidth 150 Mbps LTE there.

I'm hoping to hear more from DisplayMate soon, but these "dual-domain pixels" due seem to be touted for improved viewing angles from the literature I see. I don't know if they contribute to the 75% better contrast ratio on the iPhone 6, though (62.5% better on iP6+). This will help the screen "pop" more and close the gap a bit with OLED.

For LTE, the iPhone 5S only supported 13 bands at most with one model. The iPhone 6 supports 20, indicating there are indeed some multiband chips in there. They only have two models listed on their website so far covering the US and a few other territories, so we'll have to see how many models Asia (China) plus Eastern Europe take.
 

chrmjenkins

macrumors 603
Original poster
Oct 29, 2007
5,325
158
MD
Anandtech has a good explanation of the dual-domain pixels:

http://www.anandtech.com/show/8509/...main-pixels-in-the-iphone-6-and-iphone-6-plus

Edit: and we have confirmation the GPU is series XT (6450 or 6650) with this tweet: https://twitter.com/gavkar/status/509572966753894400

Edit 2: There were some rumors of 1810 mAh batteries and 2100 mAh batteries for the iPhone 6. Here's some calculations:

For iPhone 6 battery size. iPhone 6+ = 2915 mAh. 16 days standy. iPhone 6 xxxx mAh. 10 days. 2915/1.6 = 1821.875 mAh. Pretty safe assumption since their standby draw should be unaffected by display size, OIS and other differentiating factors.
 
Last edited:

chrmjenkins

macrumors 603
Original poster
Oct 29, 2007
5,325
158
MD
Chipworks' teardown is pretty good, and has much more techincal detail than iFixIt's.

It was nice to see the envelope tracker confirmed because it was a small part to guess at, but it also means better performance when using the cellular radio.

DisplayMate's iphone 6 and iphone 6+ display reviews are out too. http://displaymate.com/iPhone6_ShootOut.htm
 
Last edited:

chrmjenkins

macrumors 603
Original poster
Oct 29, 2007
5,325
158
MD
Chipworks has annotated their die shot. Their conclusion is quad core GPU (GX6450), which I agree with. SRAM seems roughly same size, so if they could double density, that'd give us 8MB.

For the CPU core, L2 cache appears to be independent now, from what I can tell. CPU block shrank pretty considerably overall, suggesting it is just tweaks to the Cyclone core.

It looks as if we can see two 1MB L2 caches, and I'd say the L3 is at least 6MB if not 8MB.

ByOwUlMCQAA2tXJ.jpg:large


edit: Anandtech says the SRAM is still 4MB. They back that up by quoting the exact SRAM cell size, so that seems confirmed. http://www.anandtech.com/show/8562/chipworks-a8
 
Last edited:

nordique

macrumors 68000
Oct 12, 2014
1,976
1,600
This was an exceptional thread. I remember reading through it and its still one of the best threads on the forum to this day. Great job, chrmjenkins!

Would be cool to have some spec comments on the iPhone original, 3G and 3GS chips too though in the next version of this thread :) Would just make it a nice "flow" to see what changed generation to generation


The A8 was a disappointment relative to my expectations, but I suppose that was more a testament to how "forward" the A7 design was.

The A4 -> A5 and A6 -> A7 were quite the leaps,

So here's hoping for a similar leap with the A9 from the A8.
 
  • Like
Reactions: NoWayBehind

chrmjenkins

macrumors 603
Original poster
Oct 29, 2007
5,325
158
MD
Gah


Gah! I'll be refreshing this forum all day long waiting for your post. ;)

Speaking seriously, I'm going to be cutting it much closer to the keynote date this year. Just way too busy with a lot of different things. I'll probably finish it this weekend (I hope).
 
  • Like
Reactions: Galdom

Damla

macrumors newbie
Mar 23, 2016
1
0
Could anyone please tell me what are the two components and their value, they flew off as I was removing the PM chip.
I shall be very grateful for your help.
 

Attachments

  • Iphone 6 plus.jpg
    Iphone 6 plus.jpg
    482.1 KB · Views: 160
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.