I seem to recall Siracusa talking about how GPU tasks were “embarrassingly parallel”, in that basically the way to make a GPU that’s twice as good generally starts with just doubling the number of cores one assigns.
So now that Apple has their M1 with 8 cores, is this one of those types of things where it’ll be relatively easy for them to (for example) make a Mac Pro with 32 cores?
Or is processor architecture significantly more complicated than that?
To me the M1 is the equivalent of an A14X (if it existed). My understanding is throwing more cores at the GPU is doable if the thermal management is there. But Apple has never done it before so we are still waiting to see innovation in terms of core count. I’ll be interested in this also.
Cores on GPUs aren’t equivalent to CPU cores. In the article below (you may have seen already), you can see the GPU is capable of running over 24,000 threads concurrently, 2.6 teraflops, etc. They also compare it to Intel’s latest offering. How they stack up against dedicated GPUs, I don’t know.
I really wish they would support Nvidia GPUs and CUDA, as a lot of scientific software can run on GPUs through CUDA. But, alas.
Still early days but I am concerned about this. The new models replace low end Intel models, the MacBook Air, the low end of the 13" MacBook Pro and Mac mini. They did not drop the high end Mini and MacBook Pro models. If Apple silicon does scale, we’ve yet to see them do it. And we’ve yet to see how they will handle discrete GPUs, or, indeed, that they will. Note that the M1 doesn’t support 10Gb Ethernet, RAM over 16GB, or more than two USB-C/Thunderbolt ports.
And I just read that you can’t use an EGPU on these either.
And also the very powerful, open and free 3D software Blender uses Cuda.
Blender is growing and has surpassed other tools in that space, but due to the lack of support hardly usable on Macs.
Why the consern? We have seen only the very first of Apple’s start in Mx processors. What do you see that makes you think they don’t scale? I think it is only smart that Apple starts with those Mac models where they are sure to show big improvemetns and thereby gain interest i.e. money. Apple has always been a carefull innovator, why change now?
For me, if a SOC isn’t shipping with more than 16 GB of RAM that’s not an issue. But if the SOC outright can’t even address more than 16 GB of RAM, that feels like a very problematic design limitation - even for a first-generation chip.
I’m not saying Apple is doomed or anything - but it feels like that’s a pretty huge deal to get sorted out very, very soon. Both the Macbook Pro and the Mini took a significant step backward in that regard.
Do we know it cannot address more than 16G? If not, why worry?
Besides i would expect this SoC design to undergo constant tweaking with possibly annually released upgrades. An address decoder change would be simple.
Again, why the concern, this is just the beginning of the road.
That’s what I’ve heard from various sources. I don’t know that there’s an actual technical document specifying it, but Apple tends to not release those sorts of things.
Just noting that their initial offerings are steps backward compared to RAM on at least one of the products that are being replaced.
I’m not a hardware engineer, but if the changes to support > 16 GB are simple as you suggest, why wouldn’t they launch the Pro laptop and the Mini with the ability to configure more than 16 GB?
Back on topic, as I understand it, GPU processing is parallel in the sense that if you point the work to a chip with 2 cores, it just splits the work in two halves and off they go.
4 cores, same thing.
Think of it like “we need to draw a picture. You draw the top half, I draw the bottom half and we stick them together.” That’ll be almost twice as fast as doing it myself.
With 4 cores, we just split it into 4 sections. Quick and easy.
CPU processes need to be done in sequence very often, or at least in a very controlled manner, so you can’t just split up the work so simply.
There is therefore an overhead at the splitting and recombining stages. If this overhead is larger than the time savings of splitting the job, it could actually slow the whole thing down.
Hence, CPU cores are limited because of the balance needed.
Of course Apple Silicon will scale. Would you think they go to all this trouble, after all they’ve learned, to paint themselves in a corner? We have the A-series of chips as proof that they keep improving their chips at a pace nobody matches in the industry.
This is just the beginning and they start by the base models, which makes complete sense to begin this transition. Expect the M2 to come next year, and I’d place my bet on spring for iMacs and 15’ MacBook Pros.
Like painting themselves into a thermal corner? 
I’m sure they plan for it to scale beautiful and I fully expect it will, but stuff happens.
Indeed! We know they have learned that lesson. And I think the strong issue has also been with Intel being repeatedly unable to meet their promised thermal envelopes, which is part of the reason why Apple has started that transition. They have insane room to grow now in terms of power and heat.
To me, it looks like the M1 is primarily designed as the MacBook Air chip. It replicates the specs of the existing Air in terms of RAM and Thunderbolt ports and it’s got a small enough thermal envelope that it can be accommodated in a fanless design. The Air is by far their best selling Mac, and the design seems to be optimized for it. Adding anything beyond what the Air needs would add to the cost of each and every one of the millions of them that Apple is going to make using this chip.
Having seemingly designed this chip around the MacBook Air, it appears that Apple decided to adapt the M1 to the 13” MacBook Pro and Mac mini. On one hand, this has entailed some limitations on RAM and TB ports. On the other, it has probably enabled them to get these machines out much earlier in the transition to Apple Silicon than if they waited until they had a chip ready that supported more memory and more Thunderbolt ports. They’ve made the choice to use the M1 and satisfy some users of the MacBook Pro and the Mac mini now, rather than wait until they have a chip capable enough to satisfy every MacBook Pro and Mac mini customer.
I think we shouldn’t assume that Apple is stupid.
My guess is that these models are being released in time for the holiday buying season.
There will always be one more thing, and that’s when we’ll see an expansion of capabilities.
Do you think that they’re going to keep the bifurcated-line strategy with the Pro and the Mini? One lower-specced “budget” machine, and one higher-specced “pro” sort of machine?
Or do you envision that maybe next year they’ll develop their more capable processor and then switch the entire Pro and Mini lines to that?
Product positioning. Nothing suggests the M1 is limited by design.
The 13" MacBook Pro has been bifurcated since 2015. This was more obvious when the 4-port machines had the touchbar and the 2-port machines did not (the “MacBook Escape”). However, even the more recent 2-port MacBook Pros with the touchbar were quite different than their 4-port cousins, using a different, less powerful Intel processor and a different internal design. Really two fundamentally different machines that happened to share the same name for marketing reasons. So, given that the 13" MacBook Pro has been bifurcated for the past 5 years I wouldn’t be surprised if Apple continues this division by offering a higher end machine that shares the MacBook Pro branding.
The bifurcation in the Mac mini line, on the other hand, is a new phenomenon. I could see Apple going a couple of different directions with the mini. They could drop the M1 powered mini for a more capable chip when one is available (probably whatever chip goes into the higher end MacBook Pros or the iMac), they could have two different spec machines (one with the M1, one with a more powerful chip), or they could just keep using the M1 and not offer anything more powerful in the mini.
On one hand, Apple doing the minimum necessary to keep the Mac mini as a viable product is always a good bet. If that’s the case the mini that went on sale yesterday may be the only one we see for the next three years or so.
On the other hand, the fact that Apple is still selling the higher end Intel mini may imply that there’s another shoe yet to drop. That could be a more powerful mini. Given the rumors of a smaller Mac Pro, it could also be that Apple will use the transition to Apple Silicon as an opportunity to make a cheaper, less powerful tower that will take the slot that’s been occupied by the “Pro” mini in recent years.
The one reason I have a PC under my desk.