Hard Truths About The Apple Silicon Replacement to the Mac Pro (2019)

[v0lume4]

I expect Apple to use the same 2019 Mac Pro case design but the only user accessible part may be an updated PCIe slots.

2013 Mac Pro "trash can" design is an indirect predecessor for the 2021 Mac Studio as the functionality are similar but condensed down to a 3.7L enclosure.

I would be surprised if Apple were to allow replaceable storage whether it be SATA or NVMe on those PCIe slots.

In truth there will be a severely limited audience as I expect it to sell at $6k or more and most I/O ports that almost all users need are already present in almost all Macs via TB3 ports that are PCIe-enabled.

As we progress from node to node the performance of workstations as desktops have trickled down to the laptop. That's why ~80% of PCs have been laptops for ~2 decades.

Failure of moving node to node induced Apple transitions in the past

- Jan 2006: PowerPC chip at 90nm to Intel chip at 65nm
- Nov 2020: Intel chip at 14nm to Apple chip 5nm

Yeah... I'm not expecting the best. Maybe Apple will pull a rabbit out its hat. Or maybe we're about to witness the death of the Mac Pro. Guess we'll see.

Is the market share of laptops really 80%? I guess I shouldn't be surprised, but actually seeing the number there took me by surprise. I mean, basically everyone I know has a laptop. I'm typing on one now. So, it makes sense. But still.

You know, what I wonder is if the Mac Pro will even be able to compete with performance. That's the real question. You can make an argument for paying all that money for a Mac Pro if the performance can't be matched elsewhere. But you look at what Intel, AMD, and nVidia are putting out right and it starts to make you wonder. These are, when it comes down to it, just scaled up mobile chips. It's rumored that Apple is axing the M2/M3 "Extreme" or whatever they'd call it. So would that mean just an M2 Ultra in the new Mac Pro?

I know this is just a synthetic benchmark and it isn't taking into account hardware acceleration for special tasks, but I saw this video of a mobile Intel CPU more than DOUBLING the performance of the M2 Max in CineBench and it made me wonder what Apple's response will be, if they have one:

 

[sam_dean]

Is the market share of laptops really 80%? I guess I shouldn't be surprised, but actually seeing the number there took me by surprise. I mean, basically everyone I know has a laptop. I'm typing on one now. So, it makes sense. But still.

You know, what I wonder is if the Mac Pro will even be able to compete with performance. That's the real question. You can make an argument for paying all that money for a Mac Pro if the performance can't be matched elsewhere. But you look at what Intel, AMD, and nVidia are putting out right and it starts to make you wonder. These are, when it comes down to it, just scaled up mobile chips. It's rumored that Apple is axing the M2/M3 "Extreme" or whatever they'd call it. So would that mean just an M2 Ultra in the new Mac Pro?

I know this is just a synthetic benchmark and it isn't taking into account hardware acceleration for special tasks, but I saw this video of a mobile Intel CPU more than DOUBLING the performance of the M2 Max in CineBench and it made me wonder what Apple's response will be, if they have one:

The context of laptops being ~80% is in terms of worldwide annual shipments at all price points from $0 to ~$100,000.

Laptops have become fast enough in the same sense that smartphones have become fast enough that's why a lot more people are smartphone-only rather than a smartphone + laptop or smartphone + laptop + desktop.

In the same sense that most households have a feature phone/smartphone than a refrigerator. This is in the scale of all countries rather than just the cities many of us personally visit.

Not just Macs that are start at $999 to ~$100,000. ~80% of all laptops shipped are priced below $999. Apple only focuses on the top ~20% of any market as the margins are always better. It also increases the brand value.

So say any PC OEM will out ship number of all units of laptops & desktops of all price points. But if you were to only group them to Apple's price points then Apple's market share balloons to a possible near ~80% of the whole PC OEM market.

Same logic is applied for their smartphone business. Apple's current cheapest iPhone sells for $429 and up to $1599. Per numerous reports in the past years Apple has ~80% of smartphone hardware profits. Because of their efficiency in the supply chain and only choosing to cater to the top ~20%.

Doing this also removes any legal hurdles of them being called a monopoly by choosing to limit themselves to only ~20% rather than ~80%.

Apple's smart to choose more margin per units sold rather than more units sold but less of a margin. This is why Android smartphones do not have Security Updates that lengthens to 7+ to 9+ years. IIRC Android commits to 1-3 years? They do not have the budget to fund Security Updates even when Google provides them.

Apple's choice to keep to as little variations on their products as possible and keeping to a single code base helps lower down the cost of Security Updates for all their devices that uses chips based on those used on the iPhone.

In the case of AMD, Intel and possibly NVidia the CPUs and GPUs they make are designed for desktops and laptop power allowance and not the limits of a smartphone battery smartphone cooling solution.

I think a 3nm M3 Extreme is a possibility because of the node used and if there is a large enough market for such a chip.

The question I have for that Intel chip is what is needed to get that raw performance

- power input requirement at the wall
- battery life in terms of hour
- performance per watt
- BTU/hr in terms of waste heat

The Intel Core i9-13980HX uses a 10nm lithography. Intel is known for providing bogus TDP. So that 157W max turbo power just for the chip may be even larger.

That 2023 MBP 16" M2 Max 5nm is limited by the 140W USB PD charger that powers the whole laptop.

That Alienware m16 laptop comes with a 330W charger. So it may throttle when it nears 330W charger output limit.

I was looking at the 2021 Mac Studio power consumption and noticed the M1 Max SKU had a max power input of 115W. That's 25W less than the 140W USB PD charger for a top-end 2021 MBP 16" M1 Max.

The M1 Ultra max power is 100W more at 225W.

It is possible when supply of M2 Ultra chips surpass demand for a 2023 Mac Studio M2 Ultra that Apple may come out with a new 2023 MBP 16" M2 Ultra SKU with a 240W USB PD charger.

 

[Chancha]

Regarding your proposed upgrade solution: That'd be one heck of an expensive RAM upgrade if you're having to purchase a whole new SoC package or expansion card (containing the SoC) -- but I don't see why it couldn't work. I wonder how the cooler would be handled... if you can simply drop a new SoC into the socket then no biggie

The term "SoC" has many interpretations but in the context of Apple Silicon, it should only refer to the wafer cut (the rainbow colored portion of Apple's PR photos). In all Apple Silicon Macs, the RAM is soldered next to the SoC on the same plane of substrate, so we can call it a package. And then the SSD NANDs are either soldered relatively further away on the board, or in Mac Studio case 1 to 2 daughter boards almost on the extreme end inside the chassis with a few NANDs on it. The storage controllers are within the SoC design so it occupies some area of the "chip". Think of it this way: a SoC simply miniturizes a traditional logic board by cutting some "chips" directly onto the "main chip". In a way an Intel's "CPUs" have been a sort of SoCs anyway, in that perspective, they have added numerous dedicated area for functions aside from just crunching numbers. With M1 Macs, there are already successful modifications that can de-solder and then soldering different density of RAMs and NANDs to upgrade or repair.

 

[leman]

Quick aside from the main convo at hand: Does the term "SoC" refer to everything on the substrate? Or does that term refer to just the M1/M2 chip itself? i.e. - The RAM chips are separate chips which I did not know -- naively -- until I saw the photos in this article: https://www.macrumors.com/2023/01/30/new-macbook-pro-features-smaller-heatsink/.

"SoC" means "system on a chip", so it usually refers to the stuff that's on the chip. A poster above suggests that the RAM is part of the SoC: this is incorrect as far the usual definition of the term goes. You can also use the term "system in a package" which would include RAM. But then again, these are just label, they don't have much meaning in themselves. Having RAM on the same package is primarily a way to reduce costs and memory consumption. That fact alone doesn't make the RAM faster.

Regarding your proposed upgrade solution: That'd be one heck of an expensive RAM upgrade if you're having to purchase a whole new SoC package or expansion card (containing the SoC) -- but I don't see why it couldn't work. I wonder how the cooler would be handled... if you can simply drop a new SoC into the socket then no biggie. If it sat on an expansion card, perhaps the expansion card could lie parallel to the motherboard, similar to NVMe drives. Then you could easily unmount/mount your cooler. That is, unless, Apple somehow sold a solution with the cooler already assembled.

My guess is that if Apple would go this way they would give you a fully assembled board with heatsink built in. Like a GPU basically. The 2019 Mac Pro relies on case airflow to do the actual cooling, and given that it's a cool concept (pun) I doubt that Apple will put an individual cooler on the SoC.

I know this is just a synthetic benchmark and it isn't taking into account hardware acceleration for special tasks, but I saw this video of a mobile Intel CPU more than DOUBLING the performance of the M2 Max in CineBench and it made me wonder what Apple's response will be, if they have one:

Apple doesn't need to have any response to this because it's all nonsense anyway. First, 13980HX is just a slightly power-limited desktop CPU. It produces these impressive scores by using more than 100 watts of power. So it's not the same hardware class, and you won't find this CPU in laptops with the same mix of features as M-series. Second, Cibenebench is a niche benchmark where x86 CPUs happen to excel at for various hardware and software reasons. When you look at a mixed benchmarks, like Geekbench, the difference is much smaller (2-3% in single-core and 30% in multi-core — still at 3-4x higher power consumption). Third, Cinebench is pretty much irrelevant as nobody uses CPU rendering in Blender or Cinema4D.

 

[Joe Dohn]

Apple doesn't need to have any response to this because it's all nonsense anyway. First, 13980HX is just a slightly power-limited desktop CPU. It produces these impressive scores by using more than 100 watts of power.

Except we're talking about the MAC PRO here, so power draw is not a priority. So yes, in this context, raw power consumption IS relevant.

If something is highly efficient but falls short on the raw speed you need, does it matter it's efficient?

 

[leman]

Except we're talking about the MAC PRO here, so power draw is not a priority. So yes, in this context, raw power consumption IS relevant.

If something is highly efficient but falls short on the raw speed you need, does it matter it's efficient?

As it usually happens on forums, multiple topics and discussions have mixed. Since the post I was replying to mentions mobile CPUs, I was talking about mobile CPUs.

For the Mac Pro to be relevant Apple needs to deliver raw performance, there is no question about it. Hopefully their next generation of chips will be more flexible in this regard.

 

[Apple Knowledge Navigator]

I believe the upcoming Mac Pro will retain the enclosure and features of 7,1 minus third-party GPU support, for the simple reason that they spent so much time and effort orchestrating internal expansion that it would make no sense to U-turn once again and essentially have a larger Mac Studio. That would be as much a middle finger as it was to introduce the trashcan and not even offer a tower anymore.

Likewise, all of the Mac Studio marketing has been targeted at prosumers - not the highest professionals like the current Mac Pro.

Also, all Apple’s products tell a story. If the upcoming Mac Pro is nothing more than a slightly more advanced Studio, then it has no reason to exist. Apple highlighted the ways that Avid audio, RED, Sonnet, Pegasus accessories and more could be configured to satisfy a number of market segments, and I stand by Apple recognising that these (admittedly niche) audiences still a Mac tower.

Finally, if Apple can’t design its own hardware that continues to support these features, then what the heck are they using to design the very products that they sell? They wouldn’t just abandon the power and flexibility that 7,1 has provided - there must be a solution. They know they can’t just dictate the market anymore like they did with the trashcan, and that it’s possible to satisfy different segments with different prices and features.

 

[IconDRT]

I tend to agree AKN, but I’m also reserving a 5% chance that Apple indeed gives the proverbial middle finger to the Mac Pro crowd or possible abandons altogether. As for what Apple uses for product design, there’s probably a server farm of 6,1s running Linux somewhere in the hills of Atherton.

 

[deconstruct60]

..
Except it isn't the whole machine. It's the CPU, GPU, and RAM. That's not a whole computer. You have storage and you have PCIe slots and I/O. The 2009, 2010, and 2012 Mac Pros did the exact same thing with a backplane. The only difference is that the GPU wasn't involved and you could upgrade the RAM separately from the processor.

.....

No. But the SSD controller is. Was on the T2 before Apple Silicon. But same distinction.

....
In the case of the iMac Pro and 2019 Mac Pro, the SSD controller was on the T2 chip. The NAND chips were on separate boards, but they are not a complete SSD on their own. The same is more or less true of all other T2 Macs (though the NAND chips are on the logic board, so from a parts standpoint, it's somewhat moot). This is also true of all Apple Silicon Macs as well as all Mac Studio models. ....

The Apple SoC die goes past just CPU and GPU and the package goes past including the RAM. As you commented to other responses, the SSD controller is on the die also. So is the Secure Element (and its accelerators ). There is a subset of the system management controller (SMC) on the die (and provision to connect to minimal boot ROM). There are the Thunderbolt controllers (off package there some TB retimers and socket PHYS support chips, but the baseline TB signal control is from inside the die). USB that Thunderbolt requires. DisplayPort output (that could be mixed into the what the normal GPU covers along with audio/video en/decoding. ) .
There is camera support. Connections to audio analog/digital conversion chip , networking (wifi/bluetooth/Ethernet via x1 PCI-e v4), and Power Management Control (PMC).

Whatever Apple does it highly likely won't diminish the chain of trust of the boot sequence they have in the current Apple Silicon systems ( modeled on the iPhones). If try to modularize the SoC package some other elements from the nearby logic board are likey to come with it (to close any easy man-in-the middle attack).

A mainstream socketed Windows SoC Package would assign some of those basic low level connection tasks ( SMC , PMC , security chip) to the also soldered to the board I/O Hub (PCH, 'Southbridge') chip.

There is pretty good chance if try to operationally 'pry' the SoC package off the main logic board that other stuff is going to pragmatically have to come along with it. Apple has entangled their main SoC die on several 'dimensions'.

 

[IconDRT]

This has been an interesting thread reading everyone's perspectives and the particular challenges Apple faces as it tries to build on the advantages of Apple Silicon while at the same time trying to deliver a platform that satisfies many/most/all of the Mac Pro user base requirements. It's sounding like the "SoC corner" is going to be a much more difficult engineering challenge than the "thermal corner" of the 6,1. Looking forward to what Apple comes up with assuming they don't simply exit the pro market.

 

[Yebubbleman]

The Apple SoC die goes past just CPU and GPU and the package goes past including the RAM. As you commented to other responses, the SSD controller is on the die also. So is the Secure Element (and its accelerators ). There is a subset of the system management controller (SMC) on the die (and provision to connect to minimal boot ROM). There are the Thunderbolt controllers (off package there some TB retimers and socket PHYS support chips, but the baseline TB signal control is from inside the die). USB that Thunderbolt requires. DisplayPort output (that could be mixed into the what the normal GPU covers along with audio/video en/decoding. ) .
There is camera support. Connections to audio analog/digital conversion chip , networking (wifi/bluetooth/Ethernet via x1 PCI-e v4), and Power Management Control (PMC).

Whatever Apple does it highly likely won't diminish the chain of trust of the boot sequence they have in the current Apple Silicon systems ( modeled on the iPhones). If try to modularize the SoC package some other elements from the nearby logic board are likey to come with it (to close any easy man-in-the middle attack).

A mainstream socketed Windows SoC Package would assign some of those basic low level connection tasks ( SMC , PMC , security chip) to the also soldered to the board I/O Hub (PCH, 'Southbridge') chip.

There is pretty good chance if try to operationally 'pry' the SoC package off the main logic board that other stuff is going to pragmatically have to come along with it. Apple has entangled their main SoC die on several 'dimensions'.

I suspect that it will be no different than the T2 on the current Mac Pro. I don't know of any additional security elements that Apple Silicon Macs have over T2 Intel Macs that would be negated by a socketed/slotted/otherwise-modular SoC. You'd still need to do a DFU restore on the machine to get it to function with everything and I'd imagine that would also ensure continuity of security.

 

[DCswitch]

I have both a Mac Studio and a 2019 Mac Pro 7,1. Does anyone know if the SSD in a 2019 Mac Pro will fit in a Mac Studio if length were NOT an issue? The physical size is not as long, but I'm wondering if the drive itself will press in properly AND if you could actually activate it in DFU mode via Apple Configurator 2?

I guess I could pop out a Mac Studio drive and try it in a Mac Pro 7,1 as that would fit length-wise, but before I do all this I'm wondering if someone already has looked into it.

If it were to match up, you could create a riser cable and attach it that way.

I apologize if someone has already covered this, but I read a lot of posts and didn't find an answer.

 

[DCswitch]

I just looked at the close up pictures of both drives. On the short side of the notch on the Mac Studio, it has 10 connector ribbons. On the Mac Pro 2019, it has 12, so they are NOT compatible. Bummer Apple, wow!

 

[deconstruct60]

Except we're talking about the MAC PRO here, so power draw is not a priority. So yes, in this context, raw power consumption IS relevant.

The Mac Pro is highly unlikely to be limited to a just a M2 Max. The M1 Ultra was about 1.95 faster than a M1 Max. If apply the same multiple to the M2 Pro then would basically see a doubling in performance with only a doubling in in power. So would get the performance substantially under the 100W power gap. And if task any GPU workload at the those two , then the 13980HX it would be 'crushed' by a M2 Ultra much worse that just a 'doubling'.

To a large extent what looking at that narrow context benchmark is a difference in core count. Approximately 'Level up' the core count on the Apple side to fill the same "Wattage consumption" level and there really isn't that much of a deficit. The volume of the Mac Pro 2019 case means that Apple wouldn't have to limit the CPU core count much. Two or four times the M2 Max for CPU cores would plenty of CPU performance range. If Apple did fall short on putting together a '4x count' package then there would be more limits at the top end. [ However, if they failed it is more likely a problematical disaggregation approach than any 'flaw' with their P and E core building blocks. If try to pound a 'square' monollithic laptop die into a 'round hole' as a 'chiplet' then probably run into issues. ]



If shift to dragging in the server market designed processors, Apple probably isn't to win the CPU core count 'war' , but they really do not need to for the Mac Pro. There is a narrow subset of the market they'll miss out on, but not the overwhelming bulk of it. Apple isn't after a unidimensional system so GPU, NPU , encode/decode AV are going to get substantive transistor budget allocations. Apple's SoC will do a broader range of tasks better than the singular Intel/AMD chips will.

If something is highly efficient but falls short on the raw speed you need, does it matter it's efficient?

Efficiency matters even at higher power. Intel Xeon SP gen 4 ( Sapphire Rapids) tops out at 60 cores short of AMD's 96 in part because it isn't as efficient. And also why 2/3 of the cores in a 13980HX are 'E' cores. The M2 Max has exactly the opposite ratio; 2/3 'P' cores. That matters. The higher the power consuming core you have the harder it is going to be to pack a larger number of them into a minimal space at higher clock speeds. That's why even down clocked the 13980 is running 100W higher. That lack of efficiency is why the clocks are slower.

Intel and AMD are behind the curve of Ampere Computing in providing CPU core count to servers. Ampere Altra Max M128 (128 core) got delivered in 2021. Those two are only going to catch up in later 2023- 2024.

 

[deconstruct60]

I just looked at the close up pictures of both drives. On the short side of the notch on the Mac Studio, it has 10 connector ribbons. On the Mac Pro 2019, it has 12, so they are NOT compatible. Bummer Apple, wow!

Really shouldn't be surprising. The Mac Pro's SSD controller is inside the T2. The T2 came out in 2017. So it highly likely uses NAND chips from the 2016-2019 era. The Mac Studio very highly likely uses an updated SSD controller ( created at least 2 years later) and NAND chips that likely do not date from the 2017-2018 era.

It is much more like trying to take the NAND chips off of a Samsung 950 and trying to put them onto 980 Pro board and trying to see if they work or not. Those SSDs were built with substantively different NAND technologies. NAND chips aren't uniform in management issues across vendors or across time or data storage format (across longer spans of time. SLC , MLC , TLC , QLC).

The "form over function" notion of 'it is shaped like an M.2 SSD so it should act like a M.2 SSD' is fundamentally flawed. The SSD modules are not SSDs. So really be a low expectation that it behaves like a SSD because it is not one.


Apple's SSD Modules are largely designed to be repair parts for the specific SSD that shipped with system originally. The 'upgrade' capacity usage is a secondary side-effect that happens to work inside the narrow confines of the primary intent. SSD Modules are not generic commodity parts. They are subcomponents of a SSD.



The other major problem with 'random mix and swap' is if Apple correct when they say the entire SSD is encrypted all the time. Doing a DFU would also erase the metadata the SSD had collected about the NAND chips health/wear for that module. Starting over from scratch when the actual wear is different from "fresh from the factory" will increasingly not work so well as the modules get older. So the whole market of buy old SSD modules and throw them into newer systems would be highly dubious even if Apple keep the NAND technology constant and the SSD controllers fully up to speed on old NAND management.

Again almost zero disconnect from the SSD modules intended usage as a repair module. A new repair module is fresh from the factory so has not appreciable wear to take into account. ( if replacing a SSD module because a NAND chip 'wore out' the last thing want to do is put in another 'worn' or 'almost worn out' module as replacement. )

 

[Joe Dohn]

Efficiency matters even at higher power. Intel Xeon SP gen 4 ( Sapphire Rapids) tops out at 60 cores short of AMD's 96 in part because it isn't as efficient. And also why 2/3 of the cores in a 13980HX are 'E' cores. The M2 Max has exactly the opposite ratio; 2/3 'P' cores. That matters. The higher the power consuming core you have the harder it is going to be to pack a larger number of them into a minimal space at higher clock speeds. That's why even down clocked the 13980 is running 100W higher. That lack of efficiency is why the clocks are slower.

But it's not what I said. I said that if something is efficient but it doesn't match the speed you need, then it's not really efficient for you.

 

[quarkysg]

But it's not what I said. I said that if something is efficient but it doesn't match the speed you need, then it's not really efficient for you.

What happens if the speed you need cannot be achieved because the design is not efficient?

As usual, nothing is pure black and white. It's always some shades of gray.

 

[falainber]

What happens if the speed you need cannot be achieved because the design is not efficient?

As usual, nothing is pure black and white. It's always some shades of gray.

That's hypothetical. The reality is that AS chips are efficient but have limited performance compared to competition. Their efficiency advantage may not live for much longer either. It has been reported that Intel's Meteor Lake will be 50% more efficient than their current processors (expected later this year).

 

[sam_dean]

That's hypothetical. The reality is that AS chips are efficient but have limited performance compared to competition. Their efficiency advantage may not live for much longer either. It has been reported that Intel's Meteor Lake will be 50% more efficient than their current processors (expected later this year).

Not achievable if they are not at a node before Apple.

They either outbid Apple or have a fab that is ahead of TSMC/Samsung.

 

[quarkysg]

That's hypothetical. The reality is that AS chips are efficient but have limited performance compared to competition. Their efficiency advantage may not live for much longer either. It has been reported that Intel's Meteor Lake will be 50% more efficient than their current processors (expected later this year).

Isn’t your reply hypothetical as well?

 

[sam_dean]

The reason why Apple has the efficiency edge is mainly due to Apple having an edge in terms of the node they use, the PDN (power delivery network) tech, and packaging.

Their ARM cores are actually more complex than the x86 competitors; significantly wider and with larger resources for out of order and speculation. Most people assume there is some kind of "magic" that makes ARM better that x86, but that is not the case. The ISA has little impact on overall power consumption given the same microarchitectural resources.

Apple uses their larger/more complex cores to their advantage, by running them at a slower clock rate. While allowing them to do more work per clock cycle. This allows them to operate on the frequency/power sweet spot for their process. One has to note that power consumption increases significantly (way higher than linear) the higher the frequency.

Here is where the PDN technology comes into play. Apple uses the most advanced technology to distribute power to keep all the functional units feed, which requires the ability to supply a lot of instantaneous power. To do so, Apple uses a 3D stacked architecture of 2 dies; one for the logic, and another one on top (or bottom depending where you look at it) to distribute the power. In contrast, almost every one else has to use the same die to do logic and distribute power.

The irony is that a simpler/smaller ARM core would have to be clocked faster in order to compete with Intel/AMD cores. And it would end up consuming the same high power.

Apple also has a very good SoC design. Meaning that they integrate most of the system on a single die; the CPUs, the GPU, the NPU (AI accelerator), the Codec (video processing), the camera block, I/O (USB, WiFi, ethernet, PCIe/TB, etc), and the memory controller.

For some stuff like AI and video encoding, having custom silicon handling it is far far more efficient than running it on a general purpose code.

Lastly, it also comes to packaging. Apple not only integrates the SoC in a single die, but it has the memory chips on the same package. This allows them to use low power mobile DDR chips, and since they are on package it also reduces significantly all the power that having the memory transactions run through the system's PCB externally would consume.

So it's a combination of Apple using a single package where Intel/AMD laptops require multiple through their PCBs to support the same functionality. As well as Apple having access to better overall fabrication technology for that single package that AMD/Intel have for theirs.

The trend seems to be that it is becoming more efficient for mobile vendors to scale up their products into laptops, than it is for desktop vendors to scale down their products into laptops.

There is also a key difference in business models: Apple is a system's vendor. Meaning that they sell the finished product, not just the processors. So they can use several parts from the vertical process to subsidize others. In this case, Apple can afford to make very good SoCs because they don't sell those chips elsewhere, meaning that they are not as pressured to make them "cheap" in terms of area for example. Since they're going to recoup the profit from elsewhere in the product.

In contrast; AMD and Intel sell their processors to OEMs, so they only get profit from the processor not the finished system. So they have to prioritize cost, by optimizing their designs for Area first and then focus on power. This is why both AMD and Intel use smaller cores, which allows them for smaller dies. But which have to be clocked faster in order to compete in performance, unfortunately that also increases power.

This is probably they key difference; Apple can afford the larger design that is more power efficient for the same performance. Whereas AMD/Intel have to aim for the smaller design that is less power efficient for the same performance.

 

[AdamNC]

Remember Optane Memory in Intel machines? Could AS end up adding something like that to the SoC's?

 

[theorist9]

Here are the fast SC and MC speeds currently listed for desktop processors on GeekBench:

Processor Benchmarks - Geekbench

If the M2 Ultra supports it, Apple could make use of the extensive cooling in the Mac Pro case to significantly increase its clock speed.

At 4.2 GHz, that should give GB SC/MC speeds of 2,300 and 35,000, exceeding the performance of both of the above. That would be notable because, at least for the past several years, there's not been a desktop processor that was no. 1 in both categories. [The workstation processors with the highest MC speeds have been slow for SC.]

4.5 GHz should give SC/MC speeds of 2,500 and 37,000.

My back-of-the-envelope math:
Current GHz/SC/MC for M2 Max in 16" MBP: 3.7/2,050/15,300
At 4.5 GHz: SC = 2,050 x 4.5/3.7 = 2,500; MC = 15,300 x 4.5/3.7 x 2* = 37,000
*Multiplying by 2 b/c it's the Ultra.

They might also be able to increase the GPU clocks, which are currently quite low.

These changes would of course give up some efficiency, but that's less of an issue for the Mac Pro than for any other Apple product—you'd still be left with CPU and GPU TDP's significantly below what the Mac Pro was designed for.

 

[sam_dean]

Apple uses die on die approach. It's more like a 2.5D than a true 3D stacking. By this I mean, The main die has the usual layout and they use the metal layers mostly for signaling. Then on the other side of the substrate they attach another thin die, with no logic. There they put most of the capacitors and they use the metal layers for power routing, which is fed through vertically into the other die (the one with the actual logic layout).


Intel is trying to do the capacitance on an extra layer on top of the metal layers for their dies.
Basically Apple gets to do away with most of the capacitors on their package, by putting them straight onto the other side of their die. This makes for a better PDN (their cores use, ironically a lot of power but for short bursts) and reduces system cost.

 

[sam_dean]

Remember Optane Memory in Intel machines? Could AS end up adding something like that to the SoC's?

In the last decade has AMD, Qualcomm, Huawei, Mediatech, NVidia or other big tech semiconductor ever have their version of it?

2023 may mean better tech has been developed beyond Intel Optane.