Next-Gen iPhone & iPhone Nano Speculation

[Pressure]

Teardown on the way by iFixit.

Internals reveal:

• Texas Instruments CD3240 driver device
• Broadcom BCM4330 802.11a/b/g/n MAC/Baseband/Radio with Integrated Bluetooth 4.0+HS and FM Transceiver
• 2 x 4Gb Elpida LP DDR2 = 1 GB DRAM in separate packages in a 64-bit configuration
• Qualcomm PM8028 Power Management IC
• Toshiba Y0A0000 Memory MCP
• Qualcomm RTR8600 multi-band/mode RF transceiver for LTE bands (under the thermal pad)
• Broadcom BCM5973 I/O controller
• Big honking 42.5Wh battery.
• Apple A5X chip

 

[ltcommander.data]

Teardown on the way by iFixit.

Internals reveal:

• Texas Instruments CD3240 driver device
• Broadcom BCM4330 802.11a/b/g/n MAC/Baseband/Radio with Integrated Bluetooth 4.0+HS and FM Transceiver
• 2 x 4Gb Elpida LP DDR2 = 1 GB DRAM in separate packages in a 64-bit configuration
• Qualcomm PM8028 Power Management IC
• Toshiba Y0A0000 Memory MCP
• Qualcomm RTR8600 multi-band/mode RF transceiver for LTE bands (under the thermal pad)
• Broadcom BCM5973 I/O controller
• Big honking 42.5Wh battery.
• Apple A5X chip

The Elpida RAM is LPDDR2-800 as demoted by the "8D" notation so RAM speed is unchanged. The A5X looks larger than the A5 relative to the NAND chip which I'm assuming hasn't changed in size. Presuming it's not just the heat spreader, that would point to the A5X remaining 45nm afterall. In which case, if the A5 was already a huge chip, I wonder how yields are with the A5X?

Did the requirement for the heat spreader prevent use of PoP for the RAM? In which case, wouldn't that make the A5X inappropriate for the next iPhone since there is less logic board space available to place separate RAM chips? That would support there being a A6, presumably 32nm, coming later this year.

EDIT: They've now removed the heat spreader and the actual A5X die now doesn't look as large, but it doesn't seem small relative the A5 either. I've asked them to take comparison measurements in the comments so hopefully they'll be a response.

 

[Exophase]

Ha, "The upgrade that earns it an 'X' is the new GPU, which Apple claims outperforms even Nvidia's Tegra 3 processor.".. iFixit must have been drinking some serious nVidia koolaid, everyone knows A5's GPU already easily beat Tegra 3's.

EDIT: They've now removed the heat spreader and the actual A5X die now doesn't look as large, but it doesn't seem small relative the A5 either. I've asked them to take comparison measurements in the comments so hopefully they'll be a response.

Is the square seen on top of the package necessarily the limits of the die size? Is it unusual for SoCs to be packaged like this (seems more like a desktop CPU than any SoCs I've seen)?

 

[ltcommander.data]

Is the square seen on top of the package necessarily the limits of the die size? Is it unusual for SoCs to be packaged like this (seems more like a desktop CPU than any SoCs I've seen)?

http://www.anandtech.com/show/5681/apples-a5x-die-and-size-revealed

Using photoshop and the mystical power of ratios we come up with a rough estimate of 10.8mm x 10.8mm for the A5X die, or 117.5mm^2. If you remember back to our iPad analysis article, we guessed that conservative scaling on a 32nm process would give Apple a ~125mm^2 die for the A5X. While there's a lot of estimation in our methodology, it appears likely that the A5X's die is built on a 28/32nm process - or at least not a 45nm process.

Anand's thinking it's a flip-chip with a 117.5mm^2 die size on 32nm.

 

[Exophase]

http://www.anandtech.com/show/5681/apples-a5x-die-and-size-revealed

Anand's thinking it's a flip-chip with a 117.5mm^2 die size on 32nm.

Thanks.

I wonder if Elpida's bankruptcy will impact Apple. Not a lot of companies were making non-PoP LPDDR2, last I checked.

 

[pcchen]

Maybe it's just temporary. Apple should be able to go back to PoP, just like they did in iPad 2.
Elpida Memory is in bad shape now. It had 481 billion yen debt, and its share is now trading at 1 ~ 3 yen per share. You can't really go lower than that because AFAIK TSE does not permit fraction share prices.

 

[benjiro]

http://www.anandtech.com/show/5681/apples-a5x-die-and-size-revealed

Anand's thinking it's a flip-chip with a 117.5mm^2 die size on 32nm.

Think there is a error, because if Apple used a 32nm version, why is the CPU clock only 1GHz? At a lower 32nm version, they are able to increase the CPU clock to 1.2 or 1.3Ghz, without any major impact on the battery life.

I can understand that they keep the GPU's clock at the same level, because they doubled the amount of GPU's. With this a double increase in power consumption, that is partially negated with a lower 32nm process.

But to leave the CPU at this state? Same problem with the memory. It seems that the GPU is limited by the bandwidth. Why not keep the GPU count the same, and do a 50% increase on the GPU's speed? Might have saved more power, the doubling the amount of cores, as they seems partially wasted?

No mater how one looks at it, the A5X really looks like a temporary solution while they still work on the A6 ( what in my opinion IS a quad core ).

Future:

The argument that Apple does more with less, can only hold out so long, because the entire market seems to slowly move to Quad Cores.

Nvidia Tegra 3 ( Cortex A9 ) 1.3Ghz Quad Core ( Available Now )
Nvidia Tegra 3 T33 ( Cortex A9 ) 1.6 Ghz Quad Core ( Next Month )
Nvidia Tegra 3 T35 ( Cortex A9 ) ... 28nm? ( Somewhere in 2012 )
Nvidia Wayne ( A15 ) Quad Core ( Somewhere in 2012/2013 )
Huawei K3V2 ( Cortex A9 ) Quad Core ( Next Month )
Qualcomm Krait Quad Core ( Q4 2012? / Q1 2013 )
TI A15 Quad Cores ( 2013? )

Funny thing is, i see people telling us that the dual core in the A5X is more then plenty. And people do not need Quad Cores. But ... when Apple releases a Quad Core, i am willing to bet, that you will people suddenly do a 180° turn, and applaud Apple for there "innovation".

Willing or not, people will need to learn to make there program's multi treated if they want the maximum power a device offers. This already holds true today, as almost every medium to high range device is a dual core ( and slowly the lower range devices are also moving into dual core ).

Anyway, that was a bit off-topic.

 

[metafor]

Honestly, with TI and Samsung's move to dual A15, I think the trend for products won't be quad until Windows on arm laptops. And even then, it's hard to argue that the metro environment provides enough use cases that are threaded enough. Perhaps if they had let legacy apps run, the extra cores could've been used for translation and virtualization overhead.

I don't see Apple moving to quad anything until an ARM based MacBook Air running OSX. Which I'm going to say is not only feasible, it's likely.

 

[Pressure]

Why bother with quad-core Cortex A9 when dual-core Cortex A15 performs better?

Quad-core is really not an argument in any way. The SGX543MP4 will totally overshadow the two extra cores.

 

[Lazy8s]

Until Apple is convinced the need for more CPU perforrmance, as they can profile in the testing of their own product, outweighs the extra power consumed, 32nm or not, they won't raise the clock of the current design. Quite a few SoCs targeted at lower cost markets still see a more optimal balance in pairing an ARM11, though highly clocked, with a comparatively modern GPU core. While a CPU can certainly be a bottleneck, that choking point may not at all follow the proportion of CPU:GPU in hgh end SoCs being sold to the open market.

If Apple have a custom ARMv7 coming, the design will obviously emphasize efficiency over performance, probably through a multicore approach looking either like nVidia's companion core or big.LITTLE.

Switching topics, the updates to several iOS apps give a little perspective on support for the new "retina" resolution. Real Racing 2 HD says it now supports the full 2048x1536 and adds 4x AA on top (!) on the new iPad, and Infinity Blade II just mentions that it "added high resolution support" for the new iPad. I'm guessing Infinity Blade II isn't rendering lower and then scaling up, but their wording definitely leaves room for that possibility.

 

[Arwin]

Didn't I read in the DF review of the Prime that the quad core GPU in there is actually five cores, with the fifth being a lower clocked one (500?) that can be used to conserve power when little power is required? Something like that may work with CPU as well? Or are they dynamically clocked like the PSP was?

 

[Laurent06]

Didn't I read in the DF review of the Prime that the quad core GPU in there is actually five cores, with the fifth being a lower clocked one (500?) that can be used to conserve power when little power is required? Something like that may work with CPU as well? Or are they dynamically clocked like the PSP was?

The prime has Tegra3 which is quad+1 core CPU not GPU.

 

[ltcommander.data]

Switching topics, the updates to several iOS apps give a little perspective on support for the new "retina" resolution. Real Racing 2 HD says it now supports the full 2048x1536 and adds 4x AA on top (!) on the new iPad, and Infinity Blade II just mentions that it "added high resolution support" for the new iPad. I'm guessing Infinity Blade II isn't rendering lower and then scaling up, but their wording definitely leaves room for that possibility.

I remember comments in one of these threads about the need for anisotropic filtering in iOS games. Maybe other games have since implemented it, but the just released Sky Gamblers: Air Supremacy is the only one I can remember that specifically mentions it's using 4x AF.

The prime has Tegra3 which is quad+1 core CPU not GPU.

What's more nVidia's 4+1 configuration is only necessary because their quad core is using TSMC 40nm G transistors in order to hit higher clock speeds in exchange for increased leakage. Their +1 core uses LP transistors to save power. Other SoC competitors on TSMC's processe use LP transistors for all their cores to begin with.

 

[anexanhume]

Chipworks says 45nm http://www.chipworks.com/en/technic...ns/2012/03/the-new-ipad-a-closer-look-inside/

 

[Pressure]

Chipworks says 45nm http://www.chipworks.com/en/technic...ns/2012/03/the-new-ipad-a-closer-look-inside/

Yeah, with a die size of 163mm^2 Apple is still not afraid to go large.

Really a shame they didn't go for 32 nm though. Guess the A6 will be quite the performance leap next year.

 

[tangey]

Based on the premise that other than graphics, nothing much else has been revealed to be different in the A5X, one can guess that most if not all of that increase in die space is the extra GPU IP. You can then extrapolate that back and make a reasonable assesment that IMG IP is around 50% of the entire A5X die, which if true is a remarkable pointer to the significance that apple places on graphics/video performance.

 

[anexanhume]

Yeah, with a die size of 163mm^2 Apple is still not afraid to go large.

Really a shame they didn't go for 32 nm though. Guess the A6 will be quite the performance leap next year.

My guess is that Samsung wasn't ready for volume on 32nm like Apple wanted. Or it could be just be that Apple wanted as few changes to the core as possible so they can focus on deploying A6 (maybe for next iPhone?)

 

[ltcommander.data]

Based on the premise that other than graphics, nothing much else has been revealed to be different in the A5X, one can guess that most if not all of that increase in die space is the extra GPU IP. You can then extrapolate that back and make a reasonable assesment that IMG IP is around 50% of the entire A5X die, which if true is a remarkable pointer to the significance that apple places on graphics/video performance.

http://www.anandtech.com/show/5685/apple-a5x-die-size-measured-16294mm2-likely-still-45nm

The GPU might even be larger, closer to 60% of die area, given Anand says the SGX543MP2 was ~30% of the A5. The overall die size makes the A5X bigger than a 12 EU dual core Sandy Bridge, 2x the size of Tegra 3. Large 45nm SoC, 40nm LTE baseband, and Retina Display really puts the huge battery yet no increase in battery life in focus. It's impressive, but a pretty brute force approach.

 

[tangey]

http://www.anandtech.com/show/5685/apple-a5x-die-size-measured-16294mm2-likely-still-45nm

The GPU might even be larger, closer to 60% of die area, given Anand says the SGX543MP2 was ~30% of the A5. The overall die size makes the A5X bigger than a 12 EU dual core Sandy Bridge, 2x the size of Tegra 3. Large 45nm SoC, 40nm LTE baseband, and Retina Display really puts the huge battery yet no increase in battery life in focus. It's impressive, but a pretty brute force approach.

I wonder is he including video encode/ decode IP in those figures. If so then although it's possible that the IP for those has been uprated in the A5X, its unlikely that they have grown to double size. Hence my estimation

 

[ltcommander.data]

I wonder is he including video encode/ decode IP in those figures. If so then although it's possible that the IP for those has been uprated in the A5X, its unlikely that they have grown to double size. Hence my estimation

http://www.ubmtechinsights.com/teardowns/new-apple-ipad-gen3-teardown-analysis/

I guess Anand was including video encode/decode or wasn't that accurate since the GPU seems to work out closer to 30% of the die according to the floor plan. It's interesting the SGX543MP4 is implemented as 2 distinct SGX543MP2 rather than one combined block. From the BOM, the iPad 3 has the highest component cost of any iPad.