Yield wouldn't be that bad - ARM CPUs are small in comparison to their x86 cousins. I would expect that you would get to 32 CPUs the same way everyone else does - build a 32 core design and blow enable fuses on bad CPUs, then badge as appropriate.
To get a desktop class GPU they will need to address memory bandwidth. I am assuming they are using a tile based rendering process so memory bandwidth is less of an issue in comparison to scan line rendering (do desktop GPUs still do that?) I would assume that they are enjoying the benefits of the system level cache with the CPU and GPU sharing. I would expect there to be some careful benchmarking of increasing GPU cores, memory speed, memory size and system cache size going on at Apple.
There isn't anything stopping Apple from supporting external GPUs, but it would require a new SoC with more PCIe lanes. External buses are much more power hungry and take up space on the die. I don't have a mental model of how much space plonking a PCIe x16 interconnect would cost in terms of area or power (taking into account you need to get that data across the chip too), but my gut reaction is that it isn't likely unless there is a customer use case they can't solve on chip.
They are running into their current node limits, so dont expect a 4Ghz+ coming unless they decide to remake the whole thing in High Performance Node. ( Which means it will cost a lot more for a relatively niche volume.)
>I wonder if they can make a 4xGPU
GPU increase should be given, and die size / yield would not be a concern / problem. The question is how will they handle the additional memory bandwidth requirement? As you can see in the Die-Shots those DDR I/O side are already take up much of the side space. Going with HBM2? Or LPDDR5?
>32xcore M1-family CPU, but I imagine the yield of such a beast would be horrible,
A 32x CPU / GPU Core M1 will only be 4x increase from its current form. If you exclude I/O part, the CPU + GPU + Cache is only a little more than 50% of die size, or ~60mm2. 4x That is only 240mm2+. The final product would be a sub 350mm2 if you include all the additional. Which is roughly inline with the current estimate of AWS Graviton 2 die size ( ~400mm2).
But yes, M1 ( A14X ) isn't really that interesting. The most exciting part is how they will go about doing high performance chip. Especially on iMac and Mac Pro.
And doubling the M1-GPU will not be enough to even surpass current Macbook Pro offerings. So they'll also need to do something more on that front.
And given that Firestorm is the fastest available CPU core, they probably don’t need 32 of them to match the highest multicore performance levels of existing AMD and Intel CPUs.
Seems like heat isn't an issue, so can they increase the clock-speed of the M1?
And will they create a larger die with 2xGPU and 2xCPUs, or will they "stitch" multiple M1 dies together on one package to have a higher yield and reduce supply chain concerns?
An M1X with 2xGPU will bring it to Radeon 5500M level which is what I have in my current 16", but still not quite to Nvidia 1080 level.
So they'll definitely need a more powerful GPU for the pro-users. I wonder if they can make a 4xGPU, or if the yield will become poor as the die size increases. Maybe they'll have to do a discrete GPU, but that doesn't seem to match their current trajectory.
And then there have been "leaks" (rumors, really) about an 32xcore M1-family CPU, but I imagine the yield of such a beast would be horrible, with all the other SOC components on the same die. Especially, if they put a 32xcore GPU on the same die.