There seems to be some anger in this response and it makes me wonder if I am being too hostile as well. Just to be clear, I enjoy writing these essays and there would be nothing to say if we agreed on everything. So thank you for taking the time to read my thoughts and respond, and if i do come off as hostile it's not my intention.
I did rewatch the WWDC Mac segment. This is what I see:
and
This is very different from what you said, that they won't use an iPad processor in the Mac. All they said is there will be Mac-specific SoCs.
Moreover, they then proudly introduce the Mac Mini with an A12Z - an iPad processor. No caveats are provided. At this point, if you want to maintain that Apple said they won't use an iPad SoC in a laptop, you really do have to provide a source.
I don't think an "expander" is necessary. The more I read about the smart connector, the more it seems like it's actually just repurposed lightning tech with constraints. It does data and power.
In other words,
the APU used in the iPad probably supports the two lightning ports on the Air without any changes and no unused lanes on either device.
By contrast, you want to use an SoC that will have four extra PCIe lanes the Air will just not use. Maybe more depending on the speeds the Thunderbolt ports use on other devices. That's OK, but it's not optimal.
So it seems there is no disagreement that the iPad-bound A14X would work fine for the Air if it stuck with its current memory capacity of up to 16GB. Perfect fit.
32GB needs a pinout, and other Macs you want the Air to share this processor with require this. You're saying that pinout would just be idle on the Air in its base configuration, or optional in manufacturing. Again, weird, but not the biggest problem with this.
The bigger problem is that if you want the Air to use 32GB, it needs more power. Everything we've seen suggests that LPDDR5 densities will top out at
16GB for the near future. This means 32GB requires you to go wider, and you need more power for those pins.
Let's get to the heart of what it is you think I am not considering. I am absolutely taking the power consumption of the APU, memory, and any active thunderbolt ports into consideration. Apple making their own silicon will not really move the ball much with regard to RAM and Thunderbolt power consumption.
The only thing you bring up that I may not be considering is the "4-6W Intel support chip." But... what is this? I am not seeing this on the
logic board. There is Apple's T2 support chip, but this is consuming a fraction of a watt. There's an Intel Thunderbolt controller, but we know that will stay, and I have already accounted for it. From what I am seeing comparing the iPhone and the Air boards, the T2 chip is the only thing that disappears because its functionality is already provided for in the APU. Maybe this gives you half a watt to play with; I think that's generous though
Can you please provide a part number for this 4-6W Intel Support chip, or link to Intel's product page for it?
Yeah, one thing that seems to get lost in these discussions about the Macbook Air is that it's a $900 machine that's supposed to be Apple's low end device and the Macbook Pro 13
exists to be the upgrade to this device. If the Air and the MBP use the same APU, and the MBP costs $500 more, what is that $500 buying the user?
We agree on every point! The MBP13 will be phased out when the 14 arrives (smaller bezels, same thermal constraints). Using roughly the same CPU configuration for the MBP14 and 16 is absolutely the right approach IMO; the distinguishing feature of the latter model being GPU cores and access to high bandwidth memory.
The iMac Pro is more of a question mark. The reason why I consider it a separate device is because I think the challenge Apple faces in creating a GPU capable of outperforming its top level devices is something it will take on in 2022. The APU I outlined in the, uh, 29.5" iMac
probably can't beat the high end Vega part. If it can, though, merging it into the iMac lineup makes sense.
I perceived hostility, and responded in the same manner.
The quote from WWDC is pretty clear to me:
"We’re designing a class of SoCs specifically for the Mac product line"
To me means that they are developing a class of SoCs specifically for the Mac. The term specifically to me means that they are for the Mac only, not iPad/iPhone SoCs. Your interpretation may be different, but this seems pretty clear cut. I believe that Apple will be using the A14 CPU and GPU cores, as well as various accelerators, in various quantities, for the Mac SoC. That they will be common is guaranteed, there is no reason to develop different CPU/GPU cores, and it leverages the R&D costs across the entire product linup. Where we differ is what you believe will be the quantity of CPU/GPU cores and the rest of what is inside the SoC. You think they are putting in 4P CPU cores, I think they are putting in 8. For the sake of this discussion, I will assume that we are both assuming 4E cores. The current top MBAir CPU can have 8 threads. Apple does not use SMT/Hyperthreading. So the new Apple SoC will need to have 8 threads, which implies 8 cores. Where we do differ is whether the 4E cores will be used to run applications. Right now, the 4E cores in the iPhone/iPad are devoted to "background" tasks, like managing the cellular connection, picking up messages in the background, etc. They do NOT run applications. I believe that the same will be true for the Mac SoC. Apple will use the 4E cores to manage things like Bluetooth, Wifi, and other background tasks as well as being active while the Mac is asleep (or in low power mode). So, if that is true, and it is an assumption on my part, with the iPhone/iPad usage as a guide, you now have an Intel 4 Core, 8 Thread i7 as the target to beat, and I believe that Apple doesn't want to beat it by a bit, they want to beat it by a lot. Therefore, 8P cores in the Mac SoC.
I also believe that Apple will structure the Mac SoC development into families. There will not be 7 or 8 Mac SoCs, and I think each Soc will be put into a variety of products. So, yes, the low end SoC will have some parts that go unused in the MBAir. This is basically silicon cost, or the direct cost of fabbing silicon with sections that go unused. This is the only cost. Apple, in the iPhone and iPad, already has the ability to shut off unused parts of the SoC when not being used. So, 2 PCIe lanes go unused, fine, shut them down and they use no power. You have a machine without off SoC RAM? Shut down the drivers for the external RAM. It still remains more econonmical to develop one SoC than a separate 4P/4E SoC for the MBAir and an 8P/4E SoC for the small MBP.
There are a number of parts shown in your link that will not be necessary. I used the desktop CPUs as my model when I wrote about the Intel Support chip, and I should not have made that assumption for a laptop chip, mistake is mine. However, getting back to the ifixit link to the logic board, I think the following ICs end up disappearing: Apple T2 chip, Intel Thunderbolt controller, Apple PMIC, possibly the Intersil chip, and the Macronix Multi I/O controller (which is basically the Intel Support chipset that I was talking about before, in all probability). More than likely, the fan can go as well. That group of parts likely consumes a few watts, and no, I don't know how many watts exactly. I can see that even if those chips are incorporated into the SoC, there will also be less need for on chip and external buffers and driver chips, as their functionality will be inside the SoC.
I have written that the AS MBAir will go down in price, probably down to $799-899, and the new small MBP (I am unsure if it stays at 13.3" or the bezels are reduced to allow for a 14" screen, so I call it the small MBP) will be a $999-$1099 unit. The small MBP will be faster (higher clock speed SoC), and have more ports, and perhaps slightly bigger screen, a better quality screen (as it is now) and active cooling. I also think the the MBP will have 4 USB4/TB4 ports, there will NOT be a 2 port MBP. This SoC also gets used on the regular Mac Mini, and possibly the small iMac.
You cannot get 32GB of RAM onto an SoC, and the low end will have 16GB RAM on SoC so you will need to provide for off SoC RAM. You allow the MBAir to have up to 32GB (maybe 24GB as another option) of RAM as a BTO. You allow the small MBP to have more than 32GB of RAM as a BTO, you pick a number (I would think 64 GB would be enough). If you are using a common SoC, it is a matter of running the pins out from the SoC on the logic board. With the number of ICs being eliminated (as I talked about above), there should be enough room, even on the current motherboard for that, and assuming that the SoC is physically bigger than the current Intel CPU.
As I have said, the Mac SoCs will be a family. The next SoC will be the Mid Range SoC. There may or may not be a small MBP using this SoC. This will be used in the MBP 16", the "big" low end iMac, and if it does come to be, the Mac Mini server version. It will have more CPU cores (12-16), more GPU cores (enough to drive 4-5K displays with good performance), and will allow an even larger amount of RAM (to at least 128GB) as BTO. I see the AS Mac 16" with a 4-5K screen. There will be clock speed variations.
The last SoC will be the High End SoC. This will be in the "big" high end iMac, and if it continues, the iMac Pro, as well as the Mac Pro. This will have more cores (24 or more), no on board RAM, no on board GPUs, and increased ML/Ai cores, or possibly more powerful ML/Ai cores. The SoC will allow for large amounts of RAM (probably 512GB), have PCIe lanes for actual PCIe slots, logic to use MPX slots, and for logic for multiple SoCs to work together. You can have 1 of these in the high end Mac, maybe with a max of 256GB, and a dGPU. If the iMac Pro coninues, there could be two of these in that machine, with perhaps a 512GB RAM limit, and 1 or 2 dGPUs. the Mac Pro would use 2-4 of the High end SoCs.
Just the way I think it will go. Only when real hardware shows up will we know how off I am.
