Apple is an existential threat to the PC

Apple's "efficiency" cores are likely as powerful as a Cortex A76 at probably much lower power consumption. Apple did not put them in the M1 just for core count (as we see in the Android world to claim "octa core". They have been content offering more meaningful 6 core configurations for a while now.
 
I see no reason why they can’t scale up from 120mm2 to 300-350mm2, like the PS5 and Xbox One Box X (I think I got that one right).

Easy, the reason is economic as I mentioned. Scaling up is only feasible if you have the sales to amortize the costs over. Which in the case of the dGPU Macs, is not a certainty.
 
Macbook Pro drove sales for their laptops in 2020, it's a very significant platform. They don't have that many necessary performance levels for their high volume platforms, in my opinion two SoCs would be plenty. For the rest they can differentiate with clocks/memory/display/storage.

That leaves the Mac Pro, which unlike the Macbook Pro might truly not have the volume ... there use a multi-SoC solution.
 
Easy, the reason is economic as I mentioned. Scaling up is only feasible if you have the sales to amortize the costs over. Which in the case of the dGPU Macs, is not a certainty.
It might happen anyway, look at the Afterburner card they make for the Mac Pro for instance, which adresses a niche within the Mac Pro niche.
I just don’t know what kind of volumes the Mac Pro has. They sold over 6 million laptops last quarter, and their total Mac volume is probably pretty well established, but I’ve seen no numbers for the Mac Pro specifically.
If it is much less than a hundred thousand systems annually, then the costs look harsh, but I’m not sure that sales actually are that low. I’m not in that market, so I can’t even stick a finger in the wind really. Numbers, anyone?

Architecturally, those might be the most intrigueing systems to speculate about, but on the other hand they are very much boxes bought to do a job. It could be argued that the balancing of feature sets, performance, cost, power draw, material choices and so on that go into making a computer for the general public is more complex. It sure as hell affects more people. In that sense, Apple started in the right end of the spectrum.
 
Addendum: Since so much of the total cost of bringing a low volume part to the market lies in fixed costs, you might as well go with really large dies. 600mm2 will still produce 90 dies per wafer, so even with modest yields, your cost per die is relatively low.

Of course, if volumes are higher (or fixed costs lower) you start having arguments for saving money by reducing die size. I’m sure Apple beancounters have been using Numbers to assess a multitude of scenarios. I hope they make a big chip for the Mac Pro, maybe dropping the average cost by using it in a large iMac as well. As usual, Apple has said nothing. How many SoCs are required to constitute the ”family of SoCs” that Johny Srouji mentioned?
 
It might happen anyway, look at the Afterburner card they make for the Mac Pro for instance, which adresses a niche within the Mac Pro niche.
I just don’t know what kind of volumes the Mac Pro has. They sold over 6 million laptops last quarter, and their total Mac volume is probably pretty well established, but I’ve seen no numbers for the Mac Pro specifically.
If it is much less than a hundred thousand systems annually, then the costs look harsh, but I’m not sure that sales actually are that low. I’m not in that market, so I can’t even stick a finger in the wind really. Numbers, anyone?
During Apple's Mac Pro apology in April 2017, they gave a wide range for the Mac Pro.
TechCrunch said:
Fifteen percent of all Mac customers, he says, use “pro” category apps multiple times a week. Basically, they use apps in a professional capacity. Thirty percent of Apple’s Mac customer base use pro apps less frequently than once a week but still do use them. And the vast majority of customers who define themselves as pros use Apple’s notebooks.

Next up, unsurprisingly, is the iMac. iMacs are being used disproportionately by pro users in recent years. Whether that’s due to the pace of updates or to the increased capability and much wider spread of the definition of “pro” is in the air. Likely a combination of both.

A “single digit percentage” — and it’s safe to assume the very low end of that single digit scale — of Apple’s Mac customers are Mac Pro customers.
 
Macbook Pro drove sales for their laptops in 2020, it's a very significant platform. They don't have that many necessary performance levels for their high volume platforms, in my opinion two SoCs would be plenty. For the rest they can differentiate with clocks/memory/display/storage.

That leaves the Mac Pro, which unlike the Macbook Pro might truly not have the volume ... there use a multi-SoC solution.

I don't know the sales figures but I'd guess a majority of those would be the Macbook Pro 13, which dosen't have a dGPU. The Macbook Pro 16 volumes I would think are a lot lower and there they offer three choices, 5300M, 5500M and 5600M. The iMac and Mac Pros offer anywhere between 560X, 580X, Vega 56/64, Pro 5500/5700, Pro Radeon VII, among others. That's a lot of different performance levels on offer and especially the highest end ones will be certainly difficult to address economically with SoCs. I'd still wager on them using dGPUs with their SoCs, whether their own or from AMD.

Mac Pro like I said could continue with desktop hardware as the market is likely too small to develop the required SoC power to match AMD and Intel's workstation class CPUs and GPUs.

It might happen anyway, look at the Afterburner card they make for the Mac Pro for instance, which adresses a niche within the Mac Pro niche.
I just don’t know what kind of volumes the Mac Pro has. They sold over 6 million laptops last quarter, and their total Mac volume is probably pretty well established, but I’ve seen no numbers for the Mac Pro specifically.
If it is much less than a hundred thousand systems annually, then the costs look harsh, but I’m not sure that sales actually are that low. I’m not in that market, so I can’t even stick a finger in the wind really. Numbers, anyone?

Architecturally, those might be the most intrigueing systems to speculate about, but on the other hand they are very much boxes bought to do a job. It could be argued that the balancing of feature sets, performance, cost, power draw, material choices and so on that go into making a computer for the general public is more complex. It sure as hell affects more people. In that sense, Apple started in the right end of the spectrum.

Similar points as above. Apple is probably going to go after the volume parts and not really invest into making the absolute highest end silicon but I could be wrong of course. If they can say dual purpose the silicon into their own datacenters it might be worth it.
 
Anandtech's review, here's GPU performance in a real -world game:

https://www.anandtech.com/show/16252/mac-mini-apple-m1-tested/3

Rise of the Tomb Raider - note this is running under emulation.

jrcsLNN.png
 
Macbook Pro drove sales for their laptops in 2020, it's a very significant platform. They don't have that many necessary performance levels for their high volume platforms, in my opinion two SoCs would be plenty. For the rest they can differentiate with clocks/memory/display/storage.

That leaves the Mac Pro, which unlike the Macbook Pro might truly not have the volume ... there use a multi-SoC solution.
The MacBook Pro 15- or 16-inch lineage has quite a lot of performance levels though, with tiers of GPU options and CPU options spanning a rather large price range. It is gonna be really interesting to see what Apple is baking for this:

- Are they going to consolidate upgrade options?
- Are they going to break the "unified memory architecture" and "Apple GPU inside" selling point so as to offer discrete GPUs in higher-end machines?
- Or are they going for more exotic multi-die setups, given the higher power envelope and the collective cost/reuse concerns for higher-end Mac products?
 
I think that's extremely unlikely.

They are going to drag developers into their little playground ... if you don't like to adjust to tiling and you're targeting Mac, you're screwed.
And that is appropriate. Speaking only as a Mac user from way back, minimum effort ports are spotted early, and met with disdain. When there are developers who do a better job, those are the ones that benefit from good word of mouth.
It is as it has always been, and IMHO, as it should be.
 
Addendum: Since so much of the total cost of bringing a low volume part to the market lies in fixed costs, you might as well go with really large dies. 600mm2 will still produce 90 dies per wafer, so even with modest yields, your cost per die is relatively low.

Of course, if volumes are higher (or fixed costs lower) you start having arguments for saving money by reducing die size. I’m sure Apple beancounters have been using Numbers to assess a multitude of scenarios. I hope they make a big chip for the Mac Pro, maybe dropping the average cost by using it in a large iMac as well. As usual, Apple has said nothing. How many SoCs are required to constitute the ”family of SoCs” that Johny Srouji mentioned?

One could argue that two is a couple and three is a family.

M1 for the MacBook Air, 13" MacBook Pro and Mac mini.
M1X for the 16" MacBook Pro, iMac and space gray Mac mini "Pro" (first half of 2021 or coinciding with WWDC).
M? for the iMac Pro and Mac Pro (late 2021 / early 2022).
 
One could argue that two is a couple and three is a family.

M1 for the MacBook Air, 13" MacBook Pro and Mac mini.
M1X for the 16" MacBook Pro, iMac and space gray Mac mini "Pro" (first half of 2021 or coinciding with WWDC).
M? for the iMac Pro and Mac Pro (late 2021 / early 2022).
This is what I’m suspecting. Which, given that the M1 is the iPad Pro A14x with a different sticker, definitely is the minimum to qualify as a "family".
Nevertheless, it may do the job. And, as MfA pointed out, with binning, salvage/cut down parts, RAM tiers and storage tiers, Apple will still be able to squeeze the last bitter drops out of their oranges.
If we assume such a line-up of SoCs as a framework for speculation, we can probably make a semi-decent guess as to the MBP 16 SoC, whereas the "desktop" chip still is pretty much completely up in the air.

While these first systems are quite nice, they are also quite conservative. The same product tiers, at the same prices, targeting the same demographic. Given the freedom rolling their own silicon provides Apple, it’s a bit ironic if "don’t rock the product matrix boat!" is what determines their line-up.
 
I think that's extremely unlikely.

They are going to drag developers into their little playground ... if you don't like to adjust to tiling and you're targeting Mac, you're screwed.

Maybe it is Apple who should serve the high-end graphics industry by designing GPUs specifically for them instead of expecting the opposite to happen with a nearly nonexistent market share ? ;)

If Apple wants to realistically avoid the pitfalls behind common antipatterns on tilers then they should make their GPUs compatible with the IMR pipeline if 99% of the industry decided that they want to optimize their software for them. Apple as it currently stands has zero bargaining power against them ...
 
Yet they deprecated OpenGL and refused to support Vulkan natively. Obviously their GPU will keep supporting immediate mode rendering with a somewhat less than ideal utilization of resources, but they have a longer term vision I think.

Lots of artists would rather just do everything on a Mac rather than have to switch to Windows for work ... slowly software developed for that captured audience will start poaching users and developers from Windows. That makes a lot more sense for Apple than committing themselves to GPUs they have no control over and which are at odds with their mainstream GPU architecture.

First their tiler will be good enough for most work even if not ideal, then important software will start targeting it as the primary platform and immediate mode GPUs will be the ones to start becoming less than ideal. Apple committing to tiling is an existential threat to immediate mode rendering too (then again, so is raytracing).
 
Last edited:
Anandtech's review, here's GPU performance in a real -world game:

https://www.anandtech.com/show/16252/mac-mini-apple-m1-tested/3

Rise of the Tomb Raider - note this is running under emulation.

jrcsLNN.png
if i use 1660 ti max-q of surface book 3 as a base(4,377.6 GFLOPS), apple M1 on mac mini has 39.6/82*4,377.6=2,114 GFLOPS. now i'm doubt 2.6TFLOPS M1 GPU is FP16 but FP32 instead, because rise of tomb raider porting from PC to MacOS in 2018 with minimum requirement nvidia gt 755m that not supported FP16 shader. @Nebuchadnezzar, could you confirm that M1 GPU 2.6TFLOPS is FP32 not FP16 with tools something like clpeak?
 
Last edited:
ups, just found the answer from manager Apple Silicon Graphics Drivers Teams.

FP16:FP32=1:1 on M1
GPU on A14 and GPU on M1 is different because A14 FP32 still half FP16 rate.

i think M1 is a beast with peak flops @ 2.6TFLOPS and peak power only 10.75W(GPU+DRAM on GFXBench Aztec High) from anandtech data.

and i guess Apple M1 GPU has relation with Imagination A-Series GPU because imgtec.com/blog said, "Most of the projects are still under wraps, but watch this space: there are some great things to come, some very soon, and some targeting 2021 and 2022".
source
 
Last edited:
Back
Top