MP 7,1 Mac Pro Apple Silicon is going to eat our lunch

[rondocap]

I'm going to be testing the M1 Max with the 32 core GPU vs a 28 core mac pro with the W6800x Duos in real world tests - mostly Red raw 8k, 6k, and some 3d work too, plus some pro res.

Im very curious what the results would be. The current M1 does poorly with 8k Red raw and heavy codecs like that, but this M1 Max with the 32 core GPU looks very promising.

If it's 90% as good in real world, that's a relative bargain vs the Mac Pro. Heck, just the W6800x Duo costs more than nearly a specced out M1 Max!

 

[goMac]

I honestly don't mind paying for insane performance, but I think having expansion ports/slots available for specialized niches is a necessity. I loved having a raid controller card in my 5,1.

I'm probably a broken record at this point, but in addition to all the "unified" features of Apple Silicon, a case that's supposed to remind people of the G4 Cube also does not point to nice big card slots for things like GPUs and larger storage modules.

I hope the big tower eventually makes its way to Apple Silicon. But it seems pretty clear it won't be happening with this Jade generation of Apple Silicon.

 

[richinaus]

I'm going to be testing the M1 Max with the 32 core GPU vs a 28 core mac pro with the W6800x Duos in real world tests - mostly Red raw 8k, 6k, and some 3d work too, plus some pro res.

Im very curious what the results would be. The current M1 does poorly with 8k Red raw and heavy codecs like that, but this M1 Max with the 32 core GPU looks very promising.

If it's 90% as good in real world, that's a relative bargain vs the Mac Pro. Heck, just the W6800x Duo costs more than nearly a specced out M1 Max!

I am looking forward to hearing your comparisons on 3D. I will be doing the exact same with a M1 Max 32core vs my 3080ti and the 5600m in my current 16". It is going to be interesting to see how it compares - and at 80% of the 3080ti to me will be more than fine.

 

[richinaus]

I'm probably a broken record at this point, but in addition to all the "unified" features of Apple Silicon, a case that's supposed to remind people of the G4 Cube also does not point to nice big card slots for things like GPUs and larger storage modules.

I hope the big tower eventually makes its way to Apple Silicon. But it seems pretty clear it won't be happening with this Jade generation of Apple Silicon.

Yep I am expecting a cube design over a tower, and tbh would far prefer this type of solution.
Ideally we get both sizes and it appeases the tower users, but all evidence points towards a more compact solution.

 

[deconstruct60]

I'm probably a broken record at this point, but in addition to all the "unified" features of Apple Silicon, a case that's supposed to remind people of the G4 Cube also does not point to nice big card slots for things like GPUs and larger storage modules.

I hope the big tower eventually makes its way to Apple Silicon. But it seems pretty clear it won't be happening with this Jade generation of Apple Silicon.

The rumor was for a "half sized" Mac Pro. That wouldn't be a Cube if just cut the tower height in half. There would still be room for several slots. The problem more so is getting PCI-e provisioning off of the Jade Dies than the case.
To get the very high aggregate bandwidth with just LPDDR5 RAM Apple has to use a 512-bit wide bus just for memory. Pragmatically that just means have less for PCI-e .

The "window" on Jade still being in play is that there doesn't appear to be any inter-die communication subsystem on the Max die. That means Jade 2C would have to be a different die. If it is a bigger die but still with 10C/32G then perhaps Apple can put something else on there. Or since they are open to two variations on 10C/32G they could be open to a tweaked 3rd.

There is nominally three x4 PCI-e provisioning the three TB . There are three x1 PCI-e v4 provisiong stuff like discrete Ethernet and SD-card reasder. If toss the TB controllers and repurpose all of that into PCI-e lane output they could either get two x8 PCI-e v4 out of that or just one x16 PCI-e v4.

With a Thunderbolt and PCI-e varaint building block they could do

[T][P]

and

[T][P]
[P][T]

For the four die set up they need a 'mirror' die anyway if trying to meet at the four-corners for interconnect.

With just x16 lanes aggreegate they can provision off of a PLEX switch to get fan out of 2-4 slots. Could use same board if the PLEX switch has two x16 input ( as the current one does). The 2c model just would be much more 'thin' on throughput over the slots. Would be capped at 3 onboard TB ports , but probably wouldn't "hurt" much. The 4c model could do 6 ports ( back to the MP 2013 count ).

What won't get is throughput for upper mid-high end add-in GPU cards to reach their full potential.
However, Apple can cut off 3rd party GPU cards without throwing away the hundreds of add-in card solutions in other areas.

The RAM capacity would also cap out at 256GB . ( where the iMac Pro stopped).

That somewhat would justify another round of Intel Mac Pro.

W-3300 would give them a system with

1. a 4TB memory cap. ( and slottable DIMMS while the "new" Mac Pro is trapped in the sub 1TB range. )
2. same 8 slots ( versus stopping at 2 maybe 4 )
3. still options for 3rd party GPU in case there is some b'reakout development among Inte or AMD


The M1 Max is in the sub 16 core W-3200 range on pure CPU work. Even if go to 20 cores , they are pretty likely going to be in the sub 32 W-3300 range on pure CPU work.

The 40 cores would cover in performance but it also drags along mandatory RAM minimums. ( 64GB at Apple prices ) and manditory GPU core buys. It may not win on costs versus a 32 core W-3300

 

[Waragainstsleep]

I wonder if Apple will take an M1 Max, maybe disable a few CPU cores and slap it in an MPX module to be used as a pure GPU. Maybe they could even make a variant of the chip with some of those CPU cores swapped out for some kind of high speed interconnect fabric and slap two of them side-by-side in a single MPX module for some kind of 64-core, 128GB beast of a GPU.

However they do it, it seems likely they'll have something more than comparable to AMD's GPUs by the time the AS Mac Pro is released. I wonder if they could introduce some kind of L2 RAM system. Maybe a dual processor M2 Max with 128GB on board RAM and up to 4 TB of L2 RAM for those who really need it.

 

[Waragainstsleep]

Yep I am expecting a cube design over a tower, and tbh would far prefer this type of solution.
Ideally we get both sizes and it appeases the tower users, but all evidence points towards a more compact solution.

I don't see them ditching the current tower after only one version. It still matches the Pro Display XDR too.

 

[LEOMODE]

Did not catch anything about eGPU.

But honestly I did not expect anything in that front either. There's no reason for them to hand out any revenue to third parties at this stage of world conquour. One needs to begratefull for eGPUs they are running now, me thinks this is the end of it for Mac with eGPUs. It will be all about integrated graphics and unified memory from now on with Macs. Only Apple SoCs for the future.

But I have to say I am interested now more about ASI SoC than what with M1 I was. If they just put the Max in the Mini with a reasonable price tag I just might have a try with that.

Interestingly, Apple Silicon still doesn't support any eGPU.

It looks like a lot of people are thinking it will be over the top more expensive. I think the other way around. I think it will be cheaper than the Intel counterpart, just like how it was priced less than Intel MacBook Air/MacBook Pros. So I expect low to mid-range to be somewhere around $8k vs $12k priced right now.

The only thing I'm concerned though is not being able to upgrade or replace any parts. In a Pro machine, I don't know how enticing that will be for professional consumers. But if they're only targeting professional consumers who have money to replace computers in a short time, and are only looking for the best performance possible, I guess the upgradability thing will be a thing of the past, given Apple Silicon will astronomically improve year by year over time.

 

[ct2k7]

Interestingly, Apple Silicon still doesn't support any eGPU.

It looks like a lot of people are thinking it will be over the top more expensive. I think the other way around. I think it will be cheaper than the Intel counterpart, just like how it was priced less than Intel MacBook Air/MacBook Pros. So I expect low to mid-range to be somewhere around $8k vs $12k priced right now.

The only thing I'm concerned though is not being able to upgrade or replace any parts. In a Pro machine, I don't know how enticing that will be for professional consumers. But if they're only targeting professional consumers who have money to replace computers in a short time, and are only looking for the best performance possible, I guess the upgradability thing will be a thing of the past, given Apple Silicon will astronomically improve year by year over time.

I'm torn because I don't need all the GPU power - all I want s 32GB RAM minimum, that means I need to choose wisely as it's likely I won't be able to change it in the future.

 

[goMac]

The rumor was for a "half sized" Mac Pro.

I would re-read it. The full rumor was an Apple Silicon Mac Pro that would remind people of the Cube, and a full tower Intel Mac Pro revision.

Exclusive | Mac Pro Cube Redesign with Apple Silicon for 2022 + Intel Updates

To replace the Mac Pro trashcan, Apple gave the Mac Pro a huge update in 2019 with the return of the tower design – and alongside this device came the Pro

www.idropnews.com

"According to Bloomberg’s Mark Gurman, the new Mac Pro Cube will look similar to the PowerMac G4 cube but would resemble a G4 Cube mixed with the current cheese grater design. Jon Prosser says it “looks like three or four Mac minis on top of each other.” This shows how Apple is returning to its roots by bringing back some classic things from the Steve Jobs era."

"For those of you who like the upgradability of the current Mac Pro, Apple will keep it in the lineup. It will receive updated internals from Intel and stick around for a while before Apple eventually gets rid of it."

What won't get is throughput for upper mid-high end add-in GPU cards to reach their full potential.
However, Apple can cut off 3rd party GPU cards without throwing away the hundreds of add-in card solutions in other areas.

MPX modules ain't happening, at least on this Mac Pro Mini. Jade 2c and 4c leaked at the same time as the MacBook Pro chips. They have integrated graphics. The debate window isn't completely closed. But if you believe all the other leaks from Gurman (which have been accurate so far), he's already said integrated graphics.

And the Mac Pro can't lose any depth in the case, otherwise full size PCIe cards and MPX cards won't fit. For that to happen, you'd have to believe that the Mac Pro is literally just going to loose half it's height and look like a giant bar of aluminum sitting on your desk. It doesn't make any sense. That rules out at least full width PCIe slots and MPX modules of any kind.

Which again, is why Apple keeping the Intel tower going makes sense. If they dropped MPX without it even surviving past the 2019 Mac Pro a lot of Mac Pro customers would be getting out their pitchforks.

Basically Apple can do both the 2019 and 2013 Mac Pro at the same time.

 

[randy85]

To me it's kind of obvious the AS Mac Pro won't take GPU cards because it won't need to. The core functions of the system (processing, graphics, memory) will be taken care of by the SoC and if it doesn't, then you're in the massive minority of users who need 1.5TB or ram or similar.

I can see it containing smaller PCI-E slots for things like audio and video I/O cards or mass storage.

Everything that generally crunches 1s and 0s will be catered for by the mega chip they have planned.

People that moan about lack of eGPU support refuse to see how hard the massive built-in GPUs will slap.

 

[mattspace]

People that moan about lack of eGPU support refuse to see how hard the massive built-in GPUs will slap.

Like a piece of celery, which has been in the bottom of fridge's crisper bin for 2 weeks.

They're crowing about the graphical performance of the top of the range M1 Pro Max WhateverBollocks as being competitive with the PS5 - the Lower performing of the current Sony / Microsoft console generation. The $5k+ computer having lower graphical performance than a ~$500 console is not a thing to celebrate.

 

[Abazigal]

The only thing I'm concerned though is not being able to upgrade or replace any parts. In a Pro machine, I don't know how enticing that will be for professional consumers. But if they're only targeting professional consumers who have money to replace computers in a short time, and are only looking for the best performance possible, I guess the upgradability thing will be a thing of the past, given Apple Silicon will astronomically improve year by year over time.

Is upgradability a “pro” feature or an “enthusiast” feature?

 

[mattspace]

Is upgradability a “pro” feature or an “enthusiast” feature?

"yes"

 

[cutterman]

This all sounds intriguing, but how is Apple going to surpass GPU performance on an SoC compared to discrete components? This raw computing power comes at a cost of power consumption and heat generation, which will be constrained on SoC. Also it's not like Nvidia and AMD don't have a ton of experience optimizing performance on their GPU's.

 

[deconstruct60]

This all sounds intriguing, but how is Apple going to surpass GPU performance on an SoC compared to discrete components? This raw computing power comes at a cost of power consumption and heat generation, which will be constrained on SoC.

There are a couple factors Apple is leveraging.

First, similar to gaming consoles they are pushing developers to optimize around their limitations. As the breadth of the Apple GPU implementations gets wider over an extended period of time that may loose some traction. [ For now, it pays much higher dividends for Apple to kick all other GPUs out of the pool so all optimization resource time is spent solely on their GPU. ]

Second, high power consumption isn't an advantage at hitting higher performance. It is bigger overhead. By shortening the distance between RAM and compute Apple gets to lower the overall power consumed. The lower bandwidth to individual RAM package pin is traded off versus going to a wider still bus and relatively bigger caches . Apple is following a "cheaper" HBM by stretching custom LPDDR packages into the HBM space.

Again though at a certain "scale out" factor they will probably stop seeing linear scaling on this.
( the 4x bigger Max GPU is only seeing 3x performance improvements over the M1. If multiply the Max 4 times then likely would get at least 1x loss for each of those too. So will get approximately 12x instead of a 16x. If Apple uses chiplets/tiles to implement it, then the scaling losses will probably increase incrementally. )

Third, Apple just chops certain things off. Apple GPUs can't do FP64. Not trying to track DX12 ultimate ( Windows/Xbox track) . Mostly ignoring Vulkan evolution. > 144Hz screens. Alternative memory implementations. Crypto mining . etc.

Also it's not like Nvidia and AMD don't have a ton of experience optimizing performance on their GPU's.

Optimizing for what though? To some extent Nvidia and AMD have driven off out into the swamp of tech porn benchmark chasing. In the enthusiast desktop add-in card market, power consumption is somewhat being turned into a "feature". Much of those optimizations have turned into beating on the same "old" games into faster frame rates without much new in approach.
Rumblings about AMD/Nvidia next gen top of the line cards are about new 400W power connectors and even bigger cooling assemblies. ( Nobody is putting these top end cards on a "budget". ).

But yes, if AMD/Nvidia shifted their optimization priorities a bit ( mobile optimized texture compression, using more leading edge process to go along with higher card prices, even bigger caches , easier memory mapping over PCI-e v4,5,6 ) . We'll see if AMD/Nvidia do a further 'split' between high end computational cards and 3D visuals cards.

Intel is a newer player that may have a better focus on mobile/embedded , single GPU. AMD still treats that as a secondary (maybe 3rd level) objective. Intel has Ponte Vecchio also so not 100% laser focused either.

For low end to 3080-6800 range Apple will probably be competitive. Higher than that will be a crap shoot ( especially if Apple is using tiles/chiplets and the others either are not or using relatively bigger dies. ) Stuff like MI100-MI200 , A100 , Ponte Vecchio probably not.

 

[deconstruct60]

....and are only looking for the best performance possible, I guess the upgradability thing will be a thing of the past, given Apple Silicon will astronomically improve year by year over time.

It will more so that scaling will be missing more so than upgradability ; at least short to intermediate term.

Apple won't do 3-4 GPUs workloads. Maybe they will cover some of the upper end "duo" but won't cover "quad".

Over a 4-6 year service lifetime then upgrade will be more of an issue.

The other big problem with be "regular capital expenditure with no financing" approach. Drop $2-3k per year into a machine swapping out this, that, and the other each 12-18 months. The adverse to putting the full investment up-front .

 

[deconstruct60]

I would re-read it. The full rumor was an Apple Silicon Mac Pro that would remind people of the Cube, and a full tower Intel Mac Pro revision.

Exclusive | Mac Pro Cube Redesign with Apple Silicon for 2022 + Intel Updates

To replace the Mac Pro trashcan, Apple gave the Mac Pro a huge update in 2019 with the return of the tower design – and alongside this device came the Pro

www.idropnews.com

"According to Bloomberg’s Mark Gurman, the new Mac Pro Cube will look similar to the PowerMac G4 cube but would resemble a G4 Cube mixed with the current cheese grater design. Jon Prosser says it “looks like three or four Mac minis on top of each other.” This shows how Apple is returning to its roots by bringing back some classic things from the Steve Jobs era."

"For those of you who like the upgradability of the current Mac Pro, Apple will keep it in the lineup. It will receive updated internals from Intel and stick around for a while before Apple eventually gets rid of it."

You are trying to muddle Gurman's reports with Prosser's. That isn't about re-reading, that is more so about trying to move the goal posts.

First, Prossers doesn't make sense. The Mini is 7.7" ( wide / deep) and 1.4" high. If wanted to "cube" that larger dimension that ratio is 5.5 ; not 4 or 3 . 3 or 4 would not look like a cube.
The only way to a approximately a cube would be shrink the footprint even more . That is relatively suspect if going to increase the port outputs of the mini ( to partially compensate for nixing all slots ) or cool a Jade4C ( probably in the > 280-320 W zone ). The notion with the "fan on the bottom" at a 90 degree angle to the cooling air flow for ~3000W probably isn't going to work well. So "stacked minis" is quite dubious.



Nov 2 2020

"... The new ‌Mac Pro‌ is said to have a design that looks like the current design, but in a more compact enclosure that's "about half the size." ... "

Apple Working on Redesigned Mac Pro With Smaller Form Factor and Apple Silicon Chip

As part of the transition to Apple Silicon, Apple is designing a new Mac Pro that has a smaller form factor, reports Bloomberg. The new Mac Pro is...

www.macrumors.com

May 2021
"... Alongside the faster, more powerful processor, the new ‌Mac Pro‌ will feature a smaller design that "could invoke nostalgia for the Power Mac G4 Cube, ... "

Bloomberg: Apple Silicon Mac Pro May Feature Up to 32 High Performance Cores, 128 Core Option for Graphics

Apple is working on an updated model of its Mac Pro that may feature an Apple silicon chip with as many as 32 high performance cores, and a 128 core...

www.macrumors.com

At a ~300W CPU the Mac Pro case is a far more sensible "starting point" for a Jade4C system. Provisioned for about that kind of size of package ( although Jade4C likely even bigger ) and thermals. The Mini as a reference design starting point is doesn't make any sense (if actually want to do a good job).

If chop the handle and feet height and just "halve" the actual body of the Mac Pro and would end up with something close to 8.5" wide and tall. From the front profile, it would be square. A cube's front profile is square. That will be a resemblance on that dimension. Off angle or fully from the side , it won't. But sitting shooting a picture of the Mac Pro a desk alongside a monitor it would be a frontal photo shot. the primary promotional shot in the demo or press release would look like a cube. ( also if they did a side-by-side shot like they did with the MP 2012 versus MP 2013 at the 2013's intro. )



To cut something in halve is only to cut along one dimension. If Apple is cutting across two dimensions then the correct descriptive word would be "quarter".


A 3-4 stack mini for a Jade2C that moves to "front to back" cooling would make for a fine Mac Mini Pro. If Apple collapses the enclosure so much that they exclude alterative internal storage provisioning that really shouldn't be labeled a Mac Pro. That's by Apple 's own criteria as a known problematical issue with the MP 2013 baseline design (paraphrased "leaned too hard on Thundebolt" ). If they squeeze off PCI-e v4+ lanes down to 12 x1 PCIe- v4 lanes they would have basically failed at their objective for "high bandwidth solutions" provisioning. Making folks who need > 64GB buy CPU and GPU cores they may not need is another repeat of the "value proposition" problem from before ( e.g., forcing buying 2 GPUs when users only wants one ).



There very well could be some OCD Apple industrial design rule that says that all literal desktop Macs can't have a 2D desktop surface footprint bigger than 8' x 8" . If present they will paint themselves into a corner again.

MPX modules ain't happening, at least on this Mac Pro Mini. Jade 2c and 4c leaked at the same time as the MacBook Pro chips. They have integrated graphics. The debate window isn't completely closed. But if you believe all the other leaks from Gurman (which have been accurate so far), he's already said integrated graphics.

but there are 100's of PCI-e cards that are not GPUs. What pausible rationale is there for 'throwing the baby out with the bath water ' ? Since can't do a 3rd party GPU have to eject all SDI pro 4K+ video capture cards? All multiple Tb U.2 drive sleds cards have to be ejected. 40GbE ... to the trash can.

Perhaps a fig leaf coverage to prove not being anticompetitive discrimination toward just GPU cards so discriminate against all of them. Sure they can trot out that figleaf but are they really accomplishing core Mac Pro goals at that point? If hypocrisy is a product objective is that really going to be the "best Mac Pro ever" ?

And the Mac Pro can't lose any depth in the case, otherwise full size PCIe cards and MPX cards won't fit. For that to happen, you'd have to believe that the Mac Pro is literally just going to loose half it's height and look like a giant bar of aluminum sitting on your desk. It doesn't make any sense. That rules out at least full width PCIe slots and MPX modules of any kind.

If you need more than one internal drive , video capture , > 10 GbE speed Ethernet, Afterburner, etc. how does that make sense? Having slots goes far, far , far beyond the AMD vs Nvidia fanboy wars banter. Or provisioning the built-in TB ports with DisplayPort.

The folks who need these and for which a large segment of these cards are stuffed into a external PCI-e card enclosure there likely zero net saving of actually desktop footprint consumed. The TB expansion box soaks up desktop planar space also. Maybe it is more a squarish than rectangle footprint consumed but the net space consumed is about the same.

Which again, is why Apple keeping the Intel tower going makes sense. If they dropped MPX without it even surviving past the 2019 Mac Pro a lot of Mac Pro customers would be getting out their pitchforks.

If Apple only does a Mac Mini Pro and not a "smaller" Mac Pro then yes, doing a feature bump on the current Mac Pro chassis / board would make much more sense and make it more likely.

Basically Apple can do both the 2019 and 2013 Mac Pro at the same time.

Apple pragmatically did a MP 2013 refresh with the iMac Pro. Same "one and only one" internal drive, approx 400W power supply , coupled CPU and GPU cooling ( single exhaust vent) , leaning heavily on Thunderbolt. Single Ethernet output. However, they had some sense to change to name on the product and put it into a different product line in the Mac line up.

Slapping a "Mac Pro mini" on what is really a "Mac Mini Pro" smacks of not being able to deliver on a full transition in 2 years and doing some P.T. Barnum misdirection to cover that up. Pretend that they Steve Jobs relality distortion field is turned on full blast and just "can't innovate my *ss" their way out of it. When all along it is the "same sh*t different day". Same painted into corner.


If Apple "has to" sell an improved Intel version alongside a a Mac Mini Pro then that is basically a tacit admission that really didn't complete the translon by the two year deadline. So why perpetrate the fraud when it is obviously transparent?


P.S. If Apple didn't like "Mac Mini Pro" just call it "Mac NeXT". The future of the Mac desktop. the Mac Cube was largely iterating on the original NeXT. Trying to be the mysterious "future shape" just to be different. Then it would be more clear that they were just walking away from the classic Mac Pro product space.


P.P.S. Forgot the other MP 2013 problematic issue that Apple outlined. Coupling more than one uneven thermally driven dies to one centralized core cooler. Not quite as uneven but they will need a non simplistic cooler to decouple 4 dies from cross contaminating each other when the workload doesn't spread out exactly evenly. ( which is shouldn't on most apps. ) .

 

[uller6]

Despite this little flight of fancy, it makes far more sense to assume Apple intends to stick to a single, highly scalable architecture. I can see the 2022 Mac Pros being the first to sport the next-gen M2 architecture, with scaled-down versions making their way into the mass-market Macs afterwards.

It now seems pretty likely that the upcoming big-screen iMacs will be offered with similar configurations as the new Mac Pros; something closer to the iMac Pro than a bigger-screen version of the 24" iMac. Doesn't seem wonderful for my budget - while I want something bigger than the 24", these days I don't need more performance than a garden-variety M1.

I think the next Mac pros will have the most complex version of the M1 yet - the M1 dual max or quad max with either 2 or 4 M1 max chips spliced together. I'm hoping the SOC packages are on daughter cards so the Mx Mac Pro can easily be upgraded to the Mx+1 dual/quad max when that's released.

I also think Apple will get on a ~3 year chip cadence to drive its hardware: announcing Mx in year 1, Mx pro/Mx max in year 2, and Mx dual max and Mx quad max in year 3. Then Mx+1 in year 4...etc.

 

[deconstruct60]

I think the next Mac pros will have the most complex version of the M1 yet - the M1 dual max or quad max with either 2 or 4 M1 max chips spliced together. I'm hoping the SOC packages are on daughter cards so the Mx Mac Pro can easily be upgraded to the Mx+1 dual/quad max when that's released.

Apple entangled system security with the SoC so it is very likely soldered down to minimize "man in the middle " attack vectors. The SSD is splt between the SoC and the NAND ( that hopefully are at least on replaceable daughter cards). Swapping out the possible NAND daughters cards is just taking away the data present in the system. (it won't even boot anymore since M1 Macs need a specific partition on the internal drive to fully boot up). Swapping out the SoC would implicitly means carrying away the keys also. Either way have to do a deep re-initializations and recovery of all the data. However, random person wander off with the keys is bigger security hole.

Daughter cards also cost more in power utilization and Apple is fanantical about Performance/Watt (at least for their own components. )

I also think Apple will get on a ~3 year chip cadence to drive its hardware: announcing Mx in year 1, Mx pro/Mx max in year 2, and Mx dual max and Mx quad max in year 3. Then Mx+1 in year 4...etc.

The digit will be the underlying generation. Probably nothing to do with "year". When the underlying core designs substantive change the digit will change.

The "Pro" , "Max" , "Max2" ? , "Max4" ? adjectives are die size adjustments with appropriate uncore changes to facilitate that.

Will the "Max2" and "Max4" iterate as fast as the no adjective M1/M2/M3 probably not?
the smaller dies are going to be easier to do on a faster schedule that can chase the latest process node improvements. They are also are much cheaper to buy (and make) so volume is higher to amortize higher cost fab process more easily and the yields are substantively better.

If Apple does monolithic 2c and 4c dies. If that is 2 x count more than "2 die/chip " even more so as bigger dies are harder to iterate on a quick pace while still being profitable.

The fab process full node improvements get better on about a 18 month cadence ( Moore's Law is slowing down a bit but still roughly . Waiting 3 years to do the M2 would mean skipping TSMC N4 for which there is little good reason to do. There are 9-14 month half node put into schedules regularly these days. There is little upside on sidelining the entry M-series for N3. Especially if have a N4 core implementation for the iPhone A series anyway. 2022 will probably see a better M2 come along.

A10X , A12X were skipping along on full node updates (closer to 18 months. doing just evens and skipping off A-series iterations. ) . The largest die M-series option(s) will likely take up that stance. the smaller ones may iterate faster some years if convenient ( have much less of a cost and/or power constraint problems.) .

 

[rondocap]

Another interesting point for the 2019 Mac Pro - regardless of what happens to the AS version, how far into the future will Apple provide MPX GPU upgrades?

So far they've been pretty good about it, albeit understandably a few months behind the PC versions. When The next gen AMD GPUs come out, should we still expect support even if an AS Mac Pro is already announced or coming out?

If not in an MPX module, how about drivers for the PC version of the GPUs? I know we don't know these answers, but if anyone wants to take a guess at it knowing Apples track record...

I think at least for the next gen AMD GPUs they likely should have an upgrade - since they're likely here in the next 2 years. Of course PCIE Gen 3 is a bit of a limitation on the current Mac Pro, but not entirely - should still be better performance on next gen.

If they do update the Intel Mac Pro next year, then definitely I think we'll also see more GPU MPX updates. If they don't, then it is anyone's guess as to what they'll do. This can be kept separate than their AS Mac Pro Line of course.

And I agree MPX GPUs won't be a thing in the next Mac Pro AS version - since Apple is really pushing their GPU performance. But I do agree as well that they should keep pcie slots for the plethora of other cards that take advantage of it.

Here's a thought: 2013 Mac Pro was designed into a thermal corner as said by Apple, right? So they did the opposite in the 2019 Mac Pro. What if for the next AS Mac Pro, instead of avoiding the mistakes of the 2013 Mac Pro, they want to revisit that type of framework and this time dominate it with their own silicon? They can then say, "ha! we finally got you!" instead of resigning in perpetuity with that bad design.

 

[4wdwrx]

This all sounds intriguing, but how is Apple going to surpass GPU performance on an SoC compared to discrete components? This raw computing power comes at a cost of power consumption and heat generation, which will be constrained on SoC. Also it's not like Nvidia and AMD don't have a ton of experience optimizing performance on their GPU's.

I think it's technological progress. AMD APUs are faster than many older dGPUs. A smart phone has more computing power than NASA's Apollo 13.


Same idea behind ARM compared to x86. More efficient architecture.


CPUs and GPUs are complicated and expensive to develop. The new products and generations we see every year, while have improvements are just iterative and traces of the original architecture.

Moore's Law show great improvement for Intel for decades, but it has reached its limit.


Apple Silicon is at its infant, so we see larger improvement between generations.

Someday it will also reach its limit. A new architecture will be introduced

 

[cutterman]

I understand what you’re saying, but it’s my understanding that Moore’s law had much to do with advances in lithography, which have probably reached near a physical limit in transistor size. No different for Apple or Intel.

One could debate the merits of Apples ARM instruction set vs Intel, but much of the performance will depend on software optimization. Certainly Apple will take advantage of this and force users and software developers into their ecosystem.

 

[Melbourne Park]

Yep, ordered a 14" and will wait until the Mini gets updated to replace my Mac Pro. The only real use I have for a stationary Mac is mostly on the storage side and to act as a content cache for the house.

This is the current Geekbench vs my Mac Pro and I don't think it makes sense to keep it in the long term.

Mac Pro (Late 2019) vs MacBookPro18,2 - Geekbench

Not sure, but evidently the 16" can cool better and thereby run a fair bit faster than the 14" Macbook Pro. You might need to check which is the best choice for yourself.

 

[kvic]

I think the next Mac pros will have the most complex version of the M1 yet - the M1 dual max or quad max with either 2 or 4 M1 max chips spliced together. I'm hoping the SOC packages are on daughter cards so the Mx Mac Pro can easily be upgraded to the Mx+1 dual/quad max when that's released.

Interestingly I had similar thought in recent days. I believe the bigger Apple silicons are likely to be "socketed" in Big iMac/Half-sized Mac Pro. In other words, the designs will embody some sort of modularity. This is mainly for easier manufacturing and better serviceability for repair. The fun for after-market upgraders will simply be a by-product.

How to slice into replaceable modules will be interesting to look forward to. Also if Apple wants to share some common parts among Big iMac, Half-sized Mac Pro (and perhaps a beefed up Mac Mini PRO).

If Apple decides modularity only for Mac Pro, one possibility is simply re-use the MPX slot. Say Half-sized Mac Pro will come with one half-length (or perhaps two-third length) MPX slot and a couple single-wide PCIe slots. The MPX module will carry the SoC and soldered RAM and cooler and come in different configurations. Within the same generation, the modules will be interchangeable. Across generations, interchangeability is a by-product; it won't be guaranteed but neither will be purposely sabotaged by design.

Apple Silicon is at its infant, so we see larger improvement between generations.

Someday it will also reach its limit. A new architecture will be introduced

Apple silicons have been in development and use for many years. The current huge leap in performance (per watt) on desktops and laptops are the exception and perhaps the greatest jump we'll see for a while.

Once Apple catapults all ARM-based product lines into orbits, the cadence will slow down and controlled to milk as much life-span and profit out of their investments. The best Apple fans could hope for is PC sides catch up quickly in laptop space and press Apple to remain ahead.