We are living some interesting times, aren't we now.
I really do hope they didn't paint themselves into a SoC corner.
I really do hope they didn't paint themselves into a SoC corner.
Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
MP 7,1 What must Apple include in a new Mac Pro that we can’t live without?
- Thread starter rondocap
- Start date
- Sort by reaction score
I’m really curious if a new Intel Mac Pro will support Bootcamp. It’s possible that Apple will support Bootcamp, but only do the bare minimum (ie, poor support) if at all.
Edit: My reasoning is that the rest of Apple’s machines don’t have Boot camp, so they probably won’t consider it a priority.
Edit: My reasoning is that the rest of Apple’s machines don’t have Boot camp, so they probably won’t consider it a priority.
An new Intel Mac Pro would still boot UEFI and have a T2 chip in it. It is extremely unlikely those two would go away.
"Bootcamp" is really three major factors.
a. support for partitioning/provisioning (and 'undo' that) a NFTS volume for Windows to sit on. And the MBR for the non mac bootloader.
also presenting this partition in the UEFI boot manager picker.
b. drivers on windows OS for the custom Mac components below.
c. boot environment. ( used to be presenting BIOS rather than EFI context for Windows. Now more mainstream UEFI (with BIOS fall backs if necessary for some components ). ). The T2 chip verifies the UEFI firmware and copies it into an area for the Intel to start running on it. Windows UEFI needs a signed certificate to boot securely... so that is part of that.
All of that as absolutely zero to do with macOS on M-series. There is no UEFI/EFI on M-series Macs (**). Apple writes no OS drivers for Windows. There is no supported ( there is hackery Apple left in place, but it isn't a supported mode) boot context other than into macOS.
Intel's chips need UEFI. There is code from Intel that Apple needs to allow the processor to use to bootstrap. It is technically possible to go through gobs of gryations around that, but probably would be on slippery slope with the rest of the PCI-e add-in cards firmware at that point also. (often they too are looking for UEFI at the very low levels. ).
The T2 chips were in part to aid in the transition. Keep the Intel/UEFI parts but start to roll out a different security model at the boot level. If Apple did another Intel (or AMD) x86_64 solution then a T2 ( or T3 .. with some security bug fixes ) would be pragmatically necessarily to straddle between were macOS was 10 years ago and now.
It is not a priority but it is also pretty cheap to just use the code they already have developed for the Mac Pro 2019. Not much more is required if use a relatively high degree of the same components for the system solutions. ( don't upgrade the Thunderbolt or USB controllers, very similar PCH SATA controller, same 10GbE controllers , even same PCI-e switch (only upgrade Slot 1 and 3 to PCI-e v4) , etc. )
(**) Virtual Machines provide their own UEFI emulation to get non macOS operating systems going. There isn't one down at the rare hardware level. ( Apple completely punts on that in their virtualizations framework. That's is work other vendors have to cover. )
Last edited:
They can make that argument.... it will belikely be superficially shallow. For ProRes RAW 8K ... sure works fine. But what if it is RedRAW 8K ? BlackMagic Raw 8K ?
The Afterburner FAQ says that the card did 6 8K ProRes RAW 29.97 framerate.
About the Afterburner accelerator card for Mac Pro (2019) - Apple Support
Apple Afterburner is an accelerator card for Mac Pro (2019), created to enhance Apple ProRes and ProRes RAW workflows for film and video professionals.
They used two Pro Vega II hooked with Infinity fabric and a 28 core CPU package.
Apple brags on the M1 Max
"... "One example of that is, thanks to the ProRes accelerators that are built into the Media Engine, M1 Max has two of those engines for hardware-accelerated encode and decode, so as a result, you can run seven streams of 8K ProRes in Final Cut Pro on a 16-inch MacBook Pro with M1 Max. That's more streams than on a 28-core Mac Pro with an Afterburner Card, so you're literally out-doing a workstation." ..."
Apple execs excited about M1 Max MacBook Pro video editing capabilities | AppleInsider
The M1 Max is a game-changing chip for video editing and graphics-heavy workflows, Apple executives have said in a podcast about the new MacBook Pro models, with the top-end chip capable of some video production feats that some Mac Pro variants cannot match.
appleinsider.com
Does 7 using two internal decoders so tops the 6. Using less power , less costs, and less space. Yes. But Scale? Not really. If went to a modern W6800X dual, upgraded the 28 core CPU package so that had double the bandwidth PCI-e v4 , and two Afterburner cards you would end up with a more flexible system. (at least on the decode. Afterburner is a bit crippled on doing encode. That is an Apple choice not a inherent feature of the card. ).
That dual W6800X would work on RedRaw and Blackmagic RAW also. So on breadth of utility.... there is limited value add. But if narrow the scope down to just ProRes RAW , then Apple's solution is a clear winner.
Similar issue to this kind of bragging about M1 Max beating a W6900X
Apple's M1 Max bests AMD Radeon Pro W6900X in Affinity GPU benchmark | AppleInsider
Apple's top-of-the-line M1 Max chip, which powers the company's new 14- and 16-inch MacBook Pros, outperforms a $6,000 AMD desktop GPU at certain tasks in Affinity Photo's benchmarking suite.
appleinsider.com
Interesting up until scope the context where the W6900X card is 50% bandwidth kneecapped because Mac Pro 2019 is capped at PCI-e v3 and the card is fully capable of PCI-e v4. So attach with a bechmark that copies lots of data back and forth (while kneecapping it) and then claim victory when the Max wins with the better internal bandwidth.
Meanwhile the Threadripper 5000 series is suppose to launch today. Folks on Windows with a 64 core Threadripper and two RTX 6900 with matching PCI-e v3 bussses roar past on high compute workloads with modest-high data swapping.
There is big upside in decompressing/decoding compressed video after get it into the package. Apple will win on small scale out for those. But that is dual edged sword. Doing all of that internal to the chip means it only scales as far as the chip. A Threripper 5000 box with 3 MI200 cards on a highly parallel FP64 will smoke even the Quad solution that Apple might come up with. Let alone when eventually the MI300 , MI400 solutions roll out.
I think the "more affordable for a narrow set of workloads" part woutlwork better for them. Mac apps on Apple codecs ... much better bang/buck . Yeah they have that. But doing the job of 6-7 cards... Probably don't if compare what is possible without the snail paced evolution of Apple PCI-e x16 bandwidth . (every once in a half a decade new Mac Pro).
The ability to increase GPU power over the life of the box. At reasonable cost. So PCIe slots.
Have to have drivers to do that and so far Apple isn't going down that path. Maybe they change course at WWDC 2022. If they do not change course then I wouldn't bet on 2023 or 24 it getting any better
They have once again doubled down on the "has to be power efficient" mantra for high speed connectivity to their dies. Quite doubtful there will be dGPUs from Apple either (Moving the M1 down to the iPad Airs is an even bigger boat anchor on their GPU evolution focus) .
I feel like the Mac Studio actually makes me more hopeful the next Mac Pro will stay at the high end and keep the "high end" modular features the 7,1 has, just possibly scaled down. Given what we've seen and how mammoth the M1 Ultra is it doesn't seem like a very cost-effective strategy to make even larger chips for even fewer people, especially since those chips are going to be a lot of expense for people who won't need it (for some people the GPU is overkill for their pro needs, for others it's just not enough at present and/or won't be in the future.)
Obviously we haven't seen signs yet that off-package RAM or GPUs are going to be supported yet, but it doesn't seem like there's a lot of room for a Mac Pro in the Mac Studio style. An M1 Max starts at $4K already. Why would they bother with the Mac Pro, versus just making another M1 chip with 256GB of RAM for another $1K or more tacked on the price, and calling it a day?
Obviously we haven't seen signs yet that off-package RAM or GPUs are going to be supported yet, but it doesn't seem like there's a lot of room for a Mac Pro in the Mac Studio style. An M1 Max starts at $4K already. Why would they bother with the Mac Pro, versus just making another M1 chip with 256GB of RAM for another $1K or more tacked on the price, and calling it a day?
Might as well get that out your head right now.
Thats kind of where I'm at. Hoping that Apple has the entire 2013 Mac Pro thing out of their system with this Mac. And Apple also said that M1 Ultra is the last chip in the M1 family, which means the Mac Pro might be getting something that's less SoC-y.
Last chip in the M1 family more likely means they are going to start rolling out M2's . Or drive a 'round peg into square hole' and try to make a M1 Ultra be the core of some weak slot provisioning Mac Pro.
There is a x4 PCI-e on the 'second' die and two unused TB ports. They could Rube Goldberg some slots and/or SATA drive ports out of that.
The notion they are going to go off and do something just for the Mac Pro only would be counter to everything they have done so far on "re-use" across products. The Ultra itself is a pragmatically yet another "re-use" of the Max. (somewhat only did Jade2C and sold it as a single for a while under disguise. )
If on M2 the. "twin" was a slight different die instead of an exact duplicate 'Max'. Trade the display port I/O units and Thunderbolt units not using anyway on Studio into some PCI-e v4 to provision slots. For application cores , GPU cores , RAM it would still be SoC-y. Just more general purpose I/O could put to use inside a larger box. (e.g., the Xeon 2100 in the iMac Pro had lots of unused PCI-e lanes off the CPU ... that was mainly driven by the enclosure it was in. They could build a. M2 Ultra that was a bit more flexible than what they had now and use that as an entry Mac Pro . And quad could have more permutations. ). Or this is all M3 when get to a process shrink so that Quad is not quite so much a crazy large package.
That "even larger chips for even fewer people" cuts both ways. If Apple were to build a SoC with a completely different design philosophy ( "more power consumer at low performance/ watt") then there are even fewer folks to pay for that decoupling from the rest of the Mac line up.
For the Ultra Apple used the exact same die as the Max. They don't really need to twin the die with exactly the same. It would not have to be radically different ( the Max <-> Pro feature gap isn't radically different).
From the Studio Ultra Keynote a couple of things.
1. First, an important point was they didn't want to change what Apple developers had already done much. ( so 'fused' the GPU to look like one. ). But there are a smaller subset of apps that do deal with multiple GPUs. Supplemental GPU with a backhaul are "know" to Metal.
2. Second, the Studio presenter said that the. Mac Pro 16 core and W5700 were the most popular Mac Pro product/component among users. So yes the. M1 (or M2) Ultra would cover them performance wise, but not everyone in a Mac Pro is making choices pure on CPU/GPU core benchmark score. Getting data in/out of storage drives and in/out of the system are important points too. Some folks who can live off a single disk will go much cheaper Studio but others have non CPU/GPU issues to address.
It is fewer people left, but also didn't get a substantive fraction of those folks either with the Studio. ( also have holes in Studio market where sold "too much GPU cores" to some and "too few CPU cores" to others. )
3. Apple used an extremely limited set of "lego blocks" to construct the Ultra. The attached 'twin' has additional ProRes decoder units , more CPU cores , and more GPUs cores. But because it is an exact twin it has an unused SSD controller. Several unused display output processors , at two unused Thunderbolt controllers , Secure Enclave , etc. It isn't a huge percentage of wasted space , but it also doesn't have to be wasted space.
4. The UltraFusion link using over 10,000 connections. So pretty good change can use half the number and get half the bandwidth.
If Apple shrank the number of. GPU cores ( like the Pro) and repurposed some of that space to two 10 CPU cores then would have a. 20 Core 16 GPU "lego block" that had substantially lower peak bandwidth demands. (could pair those two up and have a. 40 Core 32 GPU Duo variant. Similar if had a die where shrunk the CPU cores down (and stripped off being able to boot as a single , no SSD, Secure Enclave , etc) and had more dominate GPU allocation coupled to a higher PCI-e v4 bundle allocation then would have a "lego block" more conducive to building slot provisioning capabilities.
Apple could use those blocks in combinations over a. "Utlra" Studio , iMac Pro , Mac Pro to generate volume.
For the quad they could drop. the one fused GPU. Just need a big enough "main" fused ( two 20 16 GPU) and then two more GPGPU focused dies that present separately but still on a "Much faster than PCI-e" connection.
In short offset, smaller "left over" Mac Pro segment by composing some parts that can be used over more than just the Mac Pro user base.
How to split that up in Generation 1 probably was an unknown. But after had a solid grip on Gen 1 doing something something broader in Gen 2 (or 3 to get a full process shrink ) would be much more tractable. "Make it work , then make it fast".
There are 100's of non-GPU card implementations that have been used in Mac Pro's over the last decade or so. Just because Apple is committed to coupling their GPUs via UltraFusion links doesn't mean the rest of the card types should be thrown out the window. Throwing out GPGPU 'compute' cards doesn't make much sense either ( while Apple said in that 2017 meeting most Mac Pro users wanted "one big GPU" they also said there were others who wanted to scale bigger. )
Furthermore the Studio is going to go into much of the same "lean too hard on Thunderbolt External boxes" as the MP 2013 and iMac Pro did. An ecosystem for these PCI-e cards is going to pop up anyway. Just more space inefficient for a substantial number of users. Doesn't have to be a "max possible wall power" (1400W ) consuming container, but something back in the 600-800W range would be around a 50% cut , but cover a substantive number of other users going to "wave bye" too otherwise. ( might even cut enough weight so only needed $199 wheels.
)Am I the only one who is slightly amused, want to wave my finger at Apple and say 'I told you so' at the magical reappearance of USB-A, HDMI, SD Card slots and of course MagSafe?
Can't innovate anymore my ass...🤣
-=Glyn=-
Can't innovate anymore my ass...🤣
-=Glyn=-
they may not have the pci-e for slots. Now as for 2 ssd controllers maybe the Studio is useing them in an raid 0 config to get more speed / space. As for the unused? TB ports on die how meny TB buses are on it?That "even larger chips for even fewer people" cuts both ways. If Apple were to build a SoC with a completely different design philosophy ( "more power consumer at low performance/ watt") then there are even fewer folks to pay for that decoupling from the rest of the Mac line up.
For the Ultra Apple used the exact same die as the Max. They don't really need to twin the die with exactly the same. It would not have to be radically different ( the Max <-> Pro feature gap isn't radically different).
From the Studio Ultra Keynote a couple of things.
1. First, an important point was they didn't want to change what Apple developers had already done much. ( so 'fused' the GPU to look like one. ). But there are a smaller subset of apps that do deal with multiple GPUs. Supplemental GPU with a backhaul are "know" to Metal.
2. Second, the Studio presenter said that the. Mac Pro 16 core and W5700 were the most popular Mac Pro product/component among users. So yes the. M1 (or M2) Ultra would cover them performance wise, but not everyone in a Mac Pro is making choices pure on CPU/GPU core benchmark score. Getting data in/out of storage drives and in/out of the system are important points too. Some folks who can live off a single disk will go much cheaper Studio but others have non CPU/GPU issues to address.
It is fewer people left, but also didn't get a substantive fraction of those folks either with the Studio. ( also have holes in Studio market where sold "too much GPU cores" to some and "too few CPU cores" to others. )
3. Apple used an extremely limited set of "lego blocks" to construct the Ultra. The attached 'twin' has additional ProRes decoder units , more CPU cores , and more GPUs cores. But because it is an exact twin it has an unused SSD controller. Several unused display output processors , at two unused Thunderbolt controllers , Secure Enclave , etc. It isn't a huge percentage of wasted space , but it also doesn't have to be wasted space.
4. The UltraFusion link using over 10,000 connections. So pretty good change can use half the number and get half the bandwidth.
If Apple shrank the number of. GPU cores ( like the Pro) and repurposed some of that space to two 10 CPU cores then would have a. 20 Core 16 GPU "lego block" that had substantially lower peak bandwidth demands. (could pair those two up and have a. 40 Core 32 GPU Duo variant. Similar if had a die where shrunk the CPU cores down (and stripped off being able to boot as a single , no SSD, Secure Enclave , etc) and had more dominate GPU allocation coupled to a higher PCI-e v4 bundle allocation then would have a "lego block" more conducive to building slot provisioning capabilities.
Apple could use those blocks in combinations over a. "Utlra" Studio , iMac Pro , Mac Pro to generate volume.
For the quad they could drop. the one fused GPU. Just need a big enough "main" fused ( two 20 16 GPU) and then two more GPGPU focused dies that present separately but still on a "Much faster than PCI-e" connection.
In short offset, smaller "left over" Mac Pro segment by composing some parts that can be used over more than just the Mac Pro user base.
How to split that up in Generation 1 probably was an unknown. But after had a solid grip on Gen 1 doing something something broader in Gen 2 (or 3 to get a full process shrink ) would be much more tractable. "Make it work , then make it fast".
There are 100's of non-GPU card implementations that have been used in Mac Pro's over the last decade or so. Just because Apple is committed to coupling their GPUs via UltraFusion links doesn't mean the rest of the card types should be thrown out the window. Throwing out GPGPU 'compute' cards doesn't make much sense either ( while Apple said in that 2017 meeting most Mac Pro users wanted "one big GPU" they also said there were others who wanted to scale bigger. )
Furthermore the Studio is going to go into much of the same "lean too hard on Thunderbolt External boxes" as the MP 2013 and iMac Pro did. An ecosystem for these PCI-e cards is going to pop up anyway. Just more space inefficient for a substantial number of users. Doesn't have to be a "max possible wall power" (1400W ) consuming container, but something back in the 600-800W range would be around a 50% cut , but cover a substantive number of other users going to "wave bye" too otherwise. ( might even cut enough weight so only needed $199 wheels.
Am I the only one who is slightly amused, want to wave my finger at Apple and say 'I told you so' at the magical reappearance of USB-A, HDMI, SD Card slots and of course MagSafe?
USB-A never did disappear off the larger iMac/Mac Pro/Mini progression ( iMac Retina 5K -> iMac Pro -> Mac Studio. MP 2009 -> 2013 -> 2019 ).
It probably isn't coming back to the iMac 24" (but had it until this recent "thin out"). Didn't come back to the laptops either. Depend upon how far they shrink the low end Mini.
HDMI off the Mac Pro? When? The HDMI on the Studio? That is probably more AppleTV and Mini parts being cheap to add than fear of "Pros" wagging their fingers at Apple.
SD Card on the non 14"/16" laptops? I'd would wait to declare victory there.
MagSafe ? Again on the lighter , smaller laptops ... we'll see.
The "I told you so" notion is going to run into problems at the soldered on RAM and GPU levels because Apple actually is delivering innovation there. The Studio is beating the "maximal modular" Mac Pro 16 core / 57000 . Yes, that is partially 'kicking sand' at older parts from 2018-19 era. The Afterburner card $2K. A Studio costs $2K and does more. There are lots of folks who are going to label that as real innovation who are focused more on performance than maximum modularity.
Then they should wait for another SoC that does for a Mac Pro product. If they just want to throw out some box with 'Mac Pro' on it just to say they are done then could just use the relatively impoverished PCI on the Ultra die even more easily that the 'dead' SSD controller.
Not really. The MBP 14"/16" marketing page reports "up to 7.4GB/s" ... same number on the Studio page. If the SSD controller supports approximately equal read/write bandwidth then it is already using something like "RAID 0" already (read/write spread out over the NAND chips being used by the SSD).
Booting macOS on RAID is something Apple has been moving away from since transitioning deeper into APFS. The pre-boot environment for M-series Macs is a micro-macOS. If the primary drive dies then the whole system is pragmatically borked. Doubling that over two drives with ZERO safety added only increasing the risk of a borked system.
Should not really want long term , valuable data archive on a RAID 0 volume either.
Apple ignored the 'built in' RAID functionality of all the Intel PCH's they used. In part beause that is external but also because it is not a best practice idea for a primary boot drive. A secondary working space drive ? Sure. But a "critical data to properly boot" drive it is not a good idea.
Additionally if all this NAND is soldered to the logic board, then it is a repair risk. More modules that could fail and wipeout the> $1,000 logic board.
( won't be surprising if the "height constrained" Studio borrows from MBP 16" design choices which include soldering down the NAND. The bulk of the height addition to the baseline Mini case is for the Fans and thermal system. ).
Appears to be 4 ( and that the MBP 14"/16" unload one which incrementally probably helps with power in addition to trading for a MagSafe port ) .
The Cube>Mac Pro 2013>Mac Studio. The problem with the MP 2013 was that there was no tower Mac Pro as well and of course that they did not update it at all.
Interestingly, Apple did not reveal any clues where the new Mac Pro will go after revealing the Mac studio. Someone looked very smug when say "but that is for another day". Preview on WWDC and release later this year of early 2023 seems reasonable.
I think the core issue is pci slots. Apple seems to be more pragmatic currently and in my opinion should have at least a few slots available. Upgradeable storage is possible (albeit with Apple drives, the drive controller is on the SoC die.) but I don’t see any possible way RAM or GPU will be upgradeable.
For the moment Apple is trying to pitch this Studio as a iMac replacement. So really
Cube > Mac Pro 2013 > iMac Pro > Mac Studio
To a large extent they "upgraded the 2013"... just to went deeper on integration.
If you go to the iMac support page
iMac - Official Apple Support
Learn how to set up and use your iMac. Find all the topics, resources, and contact options you need for your iMac.
support.apple.com
and then launch to the "iMac tech specs" page ... Apple is presenting the Studio alongside the iMacs. Effectively flashing neon lights saying "transition done with iMac".
[ In some sense, they are herding folks into buying the Studio Display ... "names match so I should buy them both right?" notion. We'll see how well that works long term if they super slow motion a larger screen iMac solution. Also "right to repair" moves aren't going to mesh with their "glue to get inside" track record. In addition to component shortages on leading edge screens they have issues to sort out. ]
The bigger problem for the MP 2013 was that there was a tower ( 2010-2012 ) that was still getting upgrade GPU , SSD , etc cards. That substantive number of users sat and squatted on the MP 2009-2013 models probably wasn't a big deal for Apple. Just holding them. If slow rolls out some more RDNA2 GPU MPX modules they can hold folks rest of the year and into 2023.
They were/are trying pretty hard to say the Studio is not the Mac Pro. They are pointing at the Intel 27" iMac because that is the system they "crossed off the list" with this introduction. If the Studio was the Mac Pro replacement then wouldn't have replaced the iMac 27". The comment about the Mac pro "Not now" is just as much to get folks to realize Apple is finished with the iMac and Mini in their viewpoint ( there is a Intel Mini dangling around. Pretty likely same colocation folks that kept MP 2013 around to the bitter end at play here. Intel specific coloc service loads to service. )
I don't see that as a "real soon now" lock in for the Mac Pro. (unless doing something cheesy. Just throw out a product with name on it just to cross it off the list more so than a real true replacement. )
If the Mac Pro is destine for Feb-March 2023 (or later) hen the Fall Mac event would be a more likely place for the Mac Pro. Apple tends to stretch as far as six months or so for "sneak peaks". Longer than that there is probably some issue that isn't fully past the engineering verification testing and Apple probably doesn't want another AirPower or "thin edge" iMac introduction.
If it is just waiting on macOS 13 to roll out in the Fall , then WWDC 2022 makes lots of sense. There will be API changes being made also that will talk about.
Forgot the iMac Pro. Fits well into the non upgradable product line. If Mac Pro “only” gets a Jade 4C, then a preview later is likely. If it gets multi M1 ultras setups requiring software adaptions for full utilization, then WWDC is likely.
Larger iMacs are likely dead due to the Studio and the newly rumored M2 Pro Mac mini, but who knows?
Jade 4C feels dead at this point. Apple said Jade2C was the last M1.
Maybe we'll see a new version of Jade 4C in the M2 line. But it feels like M1 might be Apple's attempt the SoCs they have to the Mac line. And maybe the Mac Pro is somewhere beyond that with something different.
There is a decent chance that "4C" is dead permanently. Apple rides at the bleeding edge of fab processes. If they just wait for TSMC N3 then doing what could have been done in four 432 mm^2 dies could get done with just two 550mm^2 dies. ( 37% shrink on average on the die total ( N5 -> N3 is suppose to be -42% reduction so being conservative) and add 12mm^2 for some additions. ) Those dies are bigger but four major wins:
1. The packaging of the RAM and the size of the interposer are substantially more affordable. Just one interposers between the two dies. The dies are smaller so package has more room for the 16 RAM packages which soak of gobs of space themselves.
Apple perhaps could lean on the super duper bleeding edge of TSMC packaging tech, but the resulting SoC would be stratospherically priced sitting at the super upper boundaries of TSMC 3D/2.5D tech.
2. 550mm^2 is a large , but not reticule busting die. Apple bragged about beating a Nvidia RTX 3090 in some narrow benchmark. That GPU is 658mm^2 big. (so would be 108mm^2 smaller; which is about the size of an A15). They would collapsing an Ultra , "3090 beater" into one chip. [ not sure they are really going to beat the 3090 on high frame rate graphics , but with a "all on one die" they have better chance in a Studio or single die Mac Pro. ]
3. They can continue to compare themselves against the more mainstream desktop line of. AMD and Intel versus the Server package derivatives that those two sell into the high end workstation market. Apple would dominate on core count in that smaller die segment. Also their likely much lower PCI-e bandwidth won't be as much of a sore issue. If all shooting for is 2-4 slots can wrangle that with just replacing a Thunderbolt controller bank with PCI-e v4/5 provisioning on the "2nd die" for the "half sized" Mac Pro.
4. If riding the N3 , N2 , 18A train for updates, then only really need a new one around every two (or so) years.
So cheaper to build , better Ultra/iMac Pro SoCs also. , better iGPU performance , and slower pace matching Mac Pro upgrade schedule use anyway.
Downsides are:
1. that it won't be a AMD Threadripper or Intel Xeon W3000 'killer SoC'. It Won't win CPU core count wars ( not that macOS supports that anyway with the current thread count cap. ). Won't win the 300-400W , 2-4 x16 PCI-e v5/6 GPU card provisioning battle. ( again no large GPU drivers anyway in macOS).
2. If TSMC hits the a major hiccup on process shrink then loose path of going to bigger chiplet/tile count. Also counting on memory getting denser and faster to keep up with the shrinkage on main chiplet/tile .
3. Bigger dies have higher yield problems. Less expensive than expensive 4C 3D packaging , but still going to be far more expensive than the mainstream CPUs from AMD/Intel. (Pricing wise will be closer to the workstation SoCs don't want to compare to. But can counter argue that mainly buying a big GPU not a big CPU. We'll see where Intel goes when they try to package a huge iGPU tile and sell it. If high enough Apple won't look so bad. ).
4. Would have to limp on the Mac Pro 2019 into 2023.
If Apple is restricting themselves to a "Half sized" Mac Pro permanently then won't be competing with the larger workstations anyway ( HP Z8 , Dell 7000 , big BoXX , etc. ). If TSMC has major hiccup it will hurt everyone AMD, Intel, etc. Apple's pricing so far (e.g., $2K to jump from Max to Ultra ) already sets the stage for another $2K for a "Mega/Mighty/Ludicrous". $4K is enough over the actual die cost to partially pay for some bad die(s) with each sell. Mid-size die defect cost recovery is already built into the pricing structure.
If Apple tells folks they have to wait longer to get a M3 Mega Mac Pro and skip over the M2 being trotted out for some other Macs then they will get some grumbles, but most will wait. N3 was suppose to ship in 2022 and it didn't. Covid over last two years . They'll get some sympathy on a schedule delay.
Also probably will not take over the "top performing" GPU crown from the 4090/7900, but not sure Apple wants to win that contest either. Fastest iGPU is probably good enough. [ Once AMD/NVidia add big internal caches also Apple probably won't keep up at the 'biggest GPU die allocation' product level, but also will be operating 100's watts lower. ]
M2 probably doesn't buy much in tears of a getting rid of the 4 chiplet/tile count. If it is just N5P it does nothing. That is mainly a clock bump (or power consumption reduction). If it is N4P/N4 a 10-20% shrink really going to move the needle on density and thermal. improvements.
I suspect that 4C was there in case the baseline Jade didn't performance as well as they hoped. When it did do well then it got chucked. It was a "plan B". Now that the full 2+ years have past and have next gen fab process ready , they don't really need a "plan B". They could have gotten some productive work done on "semi-discrete" GPU using the connector to link a GPU treated as non unified. ( if hiccup on N3 -> N2 or N2 -> 18A ... may need to roll out a 4C solution at some point. )
I doubt Apple will go past 48 P core count over time. The fab process shrinks will get more GPU/Raytracing/NPU/Image/Video allocations on computational logic and more cache for everyone all around rather than trying to build a CPU skewed die. P cores will get a bigger budget for the area they have but not too much more area. (they'll get faster and get a bit more bandwidth as the internal data distribution bus gets more bandwidth to hand out. )
I wouldn't be surprised if the M2 series just stops at the M2 Pro level and the M3's roll out in "reverse' order. Mega , Ultra(as a single), Max/Pro , and then M3 .
Last edited:
I think there is a path to 40/128 CPU/GPU cores, but just not with four chips. And that the Mac Pro will wait for that instead. That will lower the costs for Apple (and boost margins because not passed along) and also be a higher perf/watt solution (which is Apple's main objective that they repeatedly explicitly state. )
Even if sticking with 4 chips then not a "M1" (Jade variant) class implementation.
I suspect that Apple's grand plan for large screen mini-LEDs is broken at the moment. Not affordable enough costs for them and volume issues anyway. If Apple can beat that component cost back into a zone of their likely then I suspect the larger screen iMac will return. They are "making lemonade out of lemons" now with what they can get in volume ( 'older' LED screen tech. ) . iMac 27" used to start at $1,799. A studio + studio display is $3,598. For the "gotta buy all Apple" folks that is about a 2x price increase. ( kind of like the 2x prince increase from MP 2013 -> 2019 ). That is juicy lemonade margins. Even a $1,099 Mini + $1,599 Studio display ($2698) is 50% increase.
Apple can lean on the cult buyers until they get the mini-LED problems sorted out and then come with an iMac (that beats the discrete combo on price and/or screen quality).
Last edited:
Can't say I'm not experiencing just a little bit of schadenfreude to see the iMac crowd, especially the "iMac is good enough, no one should REALLY need a Mac Pro" crowd getting reamed by Apple for a change.
What must Apple include in a new Mac Pro that we can’t live without?
128 CPU cores40 TB SSDs
But a MacBook Pro with M2 Ultra would be also nice to have.
macOS doesn't support anywhere near that number of application threads (max 64). The rest of the Mac line up and iOS/iPad OS devices don't need more than 64 either. Apple forking macOS just for one, extremely low volume Mac product is pretty unlikely. If that is a hard core requirement might as well jump onto a Windows/Linux dual Xeon SP or Epyc box at the end of 2022 (or Threadripper 5000 ).
Two 16TB U.2 drive will get to 32TB now with a single x16 PCI-e slot now.
Fusion Dual U.2 SSD PCIe Card - Tech Specs - Sonnet
View videos and tech specs for Fusion Dual U.2 SSD PCIe Card, and determine hardware and OS compatibility.
sonnet.webflow.io
Two slots and four drives would be 64TB. ( more later when higher capacity U.2 rolls out).
Apple doesn't need to provision the SSD themselves. They need to provision the slot(s) that can provision high capacity. At that point there will be multiple ways over time to provide more higher than "average speed" internal storage.
( The part should be nervous about now on Apple's SSD drives is if they are going to abandon the NAND Data cards (SSD's modules) or push soldered all the way down up to the Mac Pro also. If there is no "big" iMac to be another consumer of these modules then Apple may go cheap route (for them) and solder them down. Bad idea for customers but more money for the Scrooge McDuck money pit under HQ. )
The major problem at the moment is they are no were near provisioning an electrical x16 connection. Let alone two x16 connections with any of the SoCs they have rolled out.
If all they have is a M1 Max with UltraFusion connector is perhaps build a PCI-e expansion module and attach that to a Max via the Fusion connector. And then claim that it isn't another M1 Soc. Or perhaps once again 'hid' something in the UltraFusion connector interposer die to get them some lane provisioning .
Even going from TSMC N5 to N3 and taking only maximum power reduction benefits is only a -30% reduction in power. So about 160-170+ W and that reduction .... 112-119 W. That still wouldn't happen. The M2 either has N5P or N4/N4P which are small fractions of that reduction. Not going to happen at M2. M2 may not even have a new Max let alone a new Ultra. M2 and M2 Pro ( with some narrow tweaks. a couple of GPU and/or E cores tossed in. And an A-series , small scale ProRes tossed into the M2. ). The will be an incrementally faster M2/M3 Max along the way in a year or two that will probably suffice.
Maybe around M5, M6, or M7 generation; you have many years to save up for that system in laptop. And only if Apple doesn't "thin out" the MBP 14/16 in the mean time. (e.g., future, faster Max at lower W allows them to trim off from the MBP without backsliding. )