New Apple Silicon Macs: MacBook Air, MacBook Pro, and Mac Mini

[shumbola]

Nice scores. My overclocked 9900K multi-core score is only in the mid-9000s in Geekbench... I bet the Mac mini and MacBook Pro with active cooling come really close to the 9900K.

Now I'm even more excited about what they will do with the iMac, iMac Pro, and Mac Pro. Maybe an M2 with 8 "big" cores and no "baby" cores?

Your expectations are too high.

Yes, Apple builds a very good things, but I'm very skeptical with respect to near future Apple products with M1.
Time will tell, of course.

 

[pastrychef]

Your expectations are too high.

Yes, Apple builds a very good things, but I'm very skeptical with respect to near future Apple products with M1.
Time will tell, of course.

I don't see why... It's much easier to scale up than it is to scale down.

 

[pastrychef]

For comparison, here are some RX 560 results:

 

[Armoured]

Now I'm even more excited about what they will do with the iMac, iMac Pro, and Mac Pro. Maybe an M2 with 8 "big" cores and no "baby" cores?

I'm guessing they will always have some baby cores. And I think it's wrong to think of the baby cores as 'hurting' performance; I think apart from power management, the other significant advantage is much less heat generation, so they can reduce the thermal throttling for the "big" cores. Depends somewhat on how costly it is to switch between them for tasks. (GPU and neural engine likewise too reduce thermal throttling of main cores, and that may be partly software/compiler)

Pure speculation: apart from more big cores, I can see where could be more useful to allow the baby cores to run at same time as the big boys, to handle 'background' or low priority tasks - assuming the need to switch between high power and low power modes is entirely different compared to mobile devices. But I don't know enough technically to say whether possible or useful.

But also possible that increasing size/performance of the GPU/neural engine is where they'll really go; there will be some point at which adding more bigboy cores doesn't bring much additional benefit.

 

[Armoured]

I don't see why... It's much easier to scale up than it is to scale down.

Why do you say this? I'm not sure it's true.

Right now they've released basically one System on a Chip with basically one CPU/GPU set. (I assume the 7/8 GPU cores is some kind of binning procedure, where some with a defective GPU core are just turned off and sold as the low-end)

So it's one chipset they're selling. I don't think getting additional performance by just adding extra cores is that easy - or more to the point, very expensive. (And may not help at all with single core performance).

Apple's done extremely well at squeezing out performance from a base chipset. They have fantastic insight into the software that's actually run, where they can improve, etc. They can produce specialised SoCs at massive volumes - but that advantage starts to go away when they move into more and more cores for lower-volume products like high-end desktops.

Sure, doubling the number of bigboy cores may help - but there's always the issue that extra cores starts to have modest gains after a point - except for somewhat specialized software. And the increased performance comes at an increasingly high price for machines that aren't produced in volumes to drive down costs. As with current macpro, the market for that becomes increasingly narrow.

I personally would guess that with these low-end machines with impressive performance (supposedly), we'll see more bifurcation between a range of systems based mostly on this system on a chip, within which differentiation will be mostly things like additional memory, more ports, and other 'features.' (Sure there will be some performance differentiation as well but less dramatic than in past, and more from things like extra memory than just raw processing power, perhaps more specialised processing units - after all the phones and ipads have signal processors, image processing, machine learning etc - than from just bigboy cores). The bottom line is, most consumer/non-specialised pro will be covered by that main chipset/SoC functionality.

The next significant leap up to macpro / imacpro - nosebleed expensive and stupid-many cores and everything else they can throw at it. But - as now - tailored to quite limited markets or need for massive expansion.

I'm kind of guessing that even coming up with a system that uses standard ddr4 is unlikely - the SoC RAM seems to be very different in performance characteristics, and mixing different memory types may require some deep plumbing alterations to the system and compilers (drivers in a different sense). And assuming they come up with SoCs that reach to 32 or maybe 64 gig, they'll decide anything beyond that is in NosebleedPro market only.

Anyway, wild speculation on my part. But overall I'm guessing that's where we'll end up - two very different market segments (Nosebleed and Everything Else), and the differentiation within those will be more modest or feature oriented, not processor cores.

Also expect apple would prefer to iterate one limited chipset and increase performance on that as quickly as possible, with the coolest new features, to get more of an upgrade cycle going again. Way more attractive to apple than minor Intel upgrades where perfomance increases modestly year on year and everyone waits.

 

[pastrychef]

Why do you say this? I'm not sure it's true.

Right now they've released basically one System on a Chip with basically one CPU/GPU set. (I assume the 7/8 GPU cores is some kind of binning procedure, where some with a defective GPU core are just turned off and sold as the low-end)

So it's one chipset they're selling. I don't think getting additional performance by just adding extra cores is that easy - or more to the point, very expensive. (And may not help at all with single core performance).

Apple's done extremely well at squeezing out performance from a base chipset. They have fantastic insight into the software that's actually run, where they can improve, etc. They can produce specialised SoCs at massive volumes - but that advantage starts to go away when they move into more and more cores for lower-volume products like high-end desktops.

Sure, doubling the number of bigboy cores may help - but there's always the issue that extra cores starts to have modest gains after a point - except for somewhat specialized software. And the increased performance comes at an increasingly high price for machines that aren't produced in volumes to drive down costs. As with current macpro, the market for that becomes increasingly narrow.

I personally would guess that with these low-end machines with impressive performance (supposedly), we'll see more bifurcation between a range of systems based mostly on this system on a chip, within which differentiation will be mostly things like additional memory, more ports, and other 'features.' (Sure there will be some performance differentiation as well but less dramatic than in past, and more from things like extra memory than just raw processing power, perhaps more specialised processing units - after all the phones and ipads have signal processors, image processing, machine learning etc - than from just bigboy cores). The bottom line is, most consumer/non-specialised pro will be covered by that main chipset/SoC functionality.

The next significant leap up to macpro / imacpro - nosebleed expensive and stupid-many cores and everything else they can throw at it. But - as now - tailored to quite limited markets or need for massive expansion.

I'm kind of guessing that even coming up with a system that uses standard ddr4 is unlikely - the SoC RAM seems to be very different in performance characteristics, and mixing different memory types may require some deep plumbing alterations to the system and compilers (drivers in a different sense). And assuming they come up with SoCs that reach to 32 or maybe 64 gig, they'll decide anything beyond that is in NosebleedPro market only.

Anyway, wild speculation on my part. But overall I'm guessing that's where we'll end up - two very different market segments (Nosebleed and Everything Else), and the differentiation within those will be more modest or feature oriented, not processor cores.

Also expect apple would prefer to iterate one limited chipset and increase performance on that as quickly as possible, with the coolest new features, to get more of an upgrade cycle going again. Way more attractive to apple than minor Intel upgrades where perfomance increases modestly year on year and everyone waits.

What I meant is, they can scale up by increasing TDP. On desktops, they can use larger heatsinks/fans. They can easily make 15 or 20W TDP SoCs. Assuming performance scales well with additional power. Intel has done this. The i7-6700K was a 91W CPU and today they have the 10900K which is a 125W CPU. Plus, once you start overclocking, those numbers get much higher.

Also, tacking on more cores should not be too difficult. Amazon already has 64 core Arm CPUs.

It's much harder to scale down. Think what Intel or AMD would have to do to reduce TDP for their CPUs to run in cellphones. AMD's approach with chiplets allows them to scale the number of cores very easily but trying to cut a 105W TDP down to 10W is not trivial.

 

[Armoured]

What I meant is, they can scale up by increasing TDP. On desktops, they can use larger heatsinks/fans. They can easily make 15 or 20W TDP SoCs. Assuming performance scales well with additional power. Intel has done this. The i7-6700K was a 91W CPU and today they have the 10900K which is a 125W CPU. Plus, once you start overclocking, those numbers get much higher.

Also, tacking on more cores should not be too difficult. Amazon already has 64 core Arm CPUs.

It's much harder to scale down. Think what Intel or AMD would have to do to reduce TDP for their CPUs to run in cellphones. AMD's approach with chiplets allows them to scale the number of cores very easily but trying to cut a 105W TDP down to 10W is not trivial.

Ah, get your point.

Agree that the approach to clocking does allow for scaling up by increasing TDP (this is one approach to binning, as I understand it - they test to see which can handle the higher speeds/heat reliably). Interesting that they provided almost zero information on frequency of this first chip. (I wonder if there's a technical reason, like other parts of the SoC don't lend themselves to this change, or just being secretive - probably the latter).

I'm not sure that apple's approach to combining lots of different functionality in a single SoC lends itself as well to just adding additional cpu cores. Part of what I'm saying is (I think) apple's specific approach may change the technical/economical hurdles on this. Of course hopefully I'm wrong and they'll start coming up with 12, 16, 24, 48 etc cores, with lots of performance and pricing differentiation.

But it's striking to me they brought out two really high-volume laptops - core products - AND their only consumer mac without a monitor, and all three use the same processor/system on a chip. Not to mention it looks like in a power/capability range that would be perfectly fine for an iMac, too. (It could be though that there are other reasons for this, like on the manufacturing/design side, and made more sense to do one barnstormer first time around and only modify it afterwards).

 

[Namok88]

If you are upgrading to Big Sur a Hackintosh is going to be limited and probably slow in comparison to even the low end M1 Mac.

I would wait for real world tests and benchmarks before making assumptions based on Apple marketing.

Anybody knows they are always exaggerated.

 

[semac]

I believe in that future. Apple is playing a big game with the new M1. We will still have a few years with our Hackintoch. The question now is to follow up and see how far it will go.

 

[nicksoph]

First benchmarks out and looking good -