Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.

avro707

macrumors 68020
Dec 13, 2010
2,263
1,654
I think Apple could focus on eGPU support. Add multiple TB5 ports and provide support in macOS. Apple should spin a TB5 device chip for eGPU's and sell it. Let the eGPU vendors worry about power and the enclosure and probably leave their device specific software to them as well.
eGPU works, but it’s also a pain in the ___, because you have another box sitting on the desk with cables hanging out of it. I’ve done that before.

That’s how I got an RX6600XT flashed to work in my 5,1 Mac Pro. EGPU connected to 7,1 using Windows and then run the tools to flash the card. Then put the EGPU away in a cupboard to get it out of the way.

It’s nicer having GPUs inside the machine.

It’s just PCI-E slots needed with connections available using the cables already available and the power supply they already have which is more than up to the task. And the case they have already which is beautiful, has lots of room inside for expansion and is extremely easy to work on - I’ve done that too.

As the owner of a very high spec Mac Pro, that’s what I’d want of a future Mac Pro, it should do the lot. No compromises or shortcuts.
 
Last edited:

ZombiePhysicist

Suspended
Original poster
May 22, 2014
2,884
2,794
I dunno. If it is true that the M4 Ultra will be a monolithic chip then Apple could do a lot to optimize silicon real estate for AI purposes if they choose to eliminate the redundant components instead of reducing chip size. Of course, there is the "i want more performance cores" crowd so Apple will have to make a decision. If it was me, it would be AI focused. That is where the puck will be.

Maybe. Don't get me wrong, I would love to be blown away. The early benchmarks are that the M4 is about 25% faster in single/multicore and about 15% faster in GPU over the M3. Not sure how that translates to an Ultra chip, much less a monolithic redesign. But my sense is that might not be enough. But would love to be wrong!
 
  • Like
Reactions: steve123

Boil

macrumors 68040
Oct 23, 2018
3,477
3,173
Stargate Command
WWDC 2024
  • M4 Mac mini
  • M4 Pro Mac mini
  • M4 Max Mac Studio
  • M4 Ultra Mac Studio
  • M4 Ultra Mac Pro
Fall 2024
  • 13" M4 MacBook Air
  • 15" M4 MacBook Air
  • 14" M4 MacBook Pro
  • 14" M4 Pro MacBook Pro
  • 14" M4 Max MacBook Pro
  • 16" M4 Pro MacBook Pro
  • 16" M4 Max MacBook Pro
Yearly SoC updates, desktops @ WWDC & laptops in the Fall

M4 Ultra will be a monolithic SoC

Extreme SoC will debut with the 2nm M5 Extreme @ WWDC 2025

M5 Extreme Mac Pro Cube @ WWDC 2025
 

H. Flower

macrumors 6502a
Jul 23, 2008
759
852
The big unknown is will or wont we get an extreme for the updated Mac Pro, and will or wont we get PCIe5 and thunderbolt 5 with it?

I fear no extreme this year. My read of all the rumors is we will get a 'holistic' ultra. Just a single big die, and not a double Max. Worse, I fear we will not get PCIe5 or thunderbolt 5.

But even if we do not get an extreme, and do 'just' get a 4ultra AND get PCIe5 and thunderbolt 5, this will definitely be a machine 'to get'.

I think a holistic ultra will get us GPU performance somewhere between an AMD 6900 and an Nvidia 4090. Sadly, by then Nvidia will likely have released the followup to the 4090, and the M4ultra will not be competitive. But, the good news is it will be making up ground, and may be "good enough" for a bigger segment of apple pro/enthusiasts.

An extreme version of the chip would change things drastically in that I think it would be well beyond a 4090, and perhaps even competitive with Nvidia's latest release, which would dramatically change the pro/enthusiast landscape.

Not to mention, the ML on the M4 is going to be nuts (assuming apple makes it more easily accessible system wide).
If we go by history, youre probably right. I was starting to get excited, but they’ll probably find a way to let us all down. lol
 
  • Like
Reactions: ZombiePhysicist

impulse462

macrumors 68020
Jun 3, 2009
2,097
2,878
  • Love
Reactions: ZombiePhysicist

impulse462

macrumors 68020
Jun 3, 2009
2,097
2,878
That, and I hope that increases the chances it includes PCIe5 and Thunderbolt5.
That, and i'm curious about the AI-related hardware stuff.

I recently completed a project using apples MLX platform (its like a hybrid numpy/pytorch framework for running ML models natively on apple silicon). honestly performance was pretty great. it would be cool to have a legit 2nd platform for ML research/training other than nvidia. i feel like itd still be locked down more than nvidia, cuz with them youre just buying a gpu, but I still think it could be useful for the community.
 
  • Like
Reactions: ZombiePhysicist

avro707

macrumors 68020
Dec 13, 2010
2,263
1,654
Will Apple instead just drop both those lines completely?

That’s what I think will happen. The users (probably small compared to phones, tablets, laptops) could use PC workstations instead.

Admittedly this is the same arguments used against the Intel Mac Pro.
 
  • Wow
Reactions: ZombiePhysicist

ZombiePhysicist

Suspended
Original poster
May 22, 2014
2,884
2,794
Will Apple instead just drop both those lines completely?

That’s what I think will happen.

I certainly hope not. Anything is possible. But I do think we will get at least one more Mac Pro. I think the M2Ultra MP was a dud and they know it too. If they put out a machine that people are asking for, and it duds, then I think they may start to question it.
 
  • Like
Reactions: MacHeritage

avro707

macrumors 68020
Dec 13, 2010
2,263
1,654
then I think they may start to question it.
We will need to see if they will take the shackles off it and make it a monster workstation like the 2019 could be or they just give the market to PCs.

But I suspect those times are over. The Studio and Silicon Mac Pro users should start planning to move to a different machine in case that becomes necessary.

It would still be interesting to imagine a new Mac Pro with a pair of monster W7900 Radeons in it or maybe even up to 4 GPUs like in some PC workstations to cater for the needs that a CPU alone isn’t suitable to do.
 

deconstruct60

macrumors G5
Mar 10, 2009
12,493
4,053
Will Apple instead just drop both those lines completely?

Both? Extremely unlikely. The Studio is the replacement for the 'big iMac" segment in the price range. Apple highly likely isn't going to abandon that. The current "iPad on stick" and Mini put too many self imposed limitations on their desktop line up in terms of cooling and internal volume capacity. ( the iMac is largely restricted to the relatively narrow height/width 'chin' of the device for SoC placement. The Mini is a bit 'too big' for plain Mn but the Mn Pro is a tighter squeeze presuming want to stick with internal power supply). The studio avoids that but doesn't diveraege too far away from the constraints on the MBP 16" chassis restrictions.

The far over $10+ K workstation , maybe. But all they would have to do is decouple the Studio Ultra submodel from the Studio product. The Studio would still be around minimally with a "Max" option even ifthe "Ultra" SoC disappeared. (which it doesn't 'have to' . M1 Ultra did OK with no Mac Pro).

Or if Apple wanted to move the Studio at different speed : MP ( M4 gen -> M6 gen -> M8 gen ) and the Studio get M-odd Max stop gap fillers between 2 years gaps and no "Ultra". Presuming the laptops were on 12 month schedule.... which may not work out either. The long term slower the "Max' die size goes the slower the MP iterating more lines up. ).


Apple's deep refresh cycle on the Mac Pro has a been 4 years ( 2009-2013) , 6 years ( 20013-2019), 4 years ( 2019-2023 ). A highly regular, two year cycle would be way faster than their established average. Taking two years to do a 'refresh update' doesn't exactly scream 'abandonment by Apple'. It is just longer, but longer has been their modus operandi for over a decade. Why is that surprising?


That’s what I think will happen. The users (probably small compared to phones, tablets, laptops) could use PC workstations instead.

The hyper modular folks leave and the old iMac class collapses? The old iMac class were not highly composed of the hyper modular folks in the first place. Hyper modular folks leaving if a big 'huff' isn't going to substantively shrink the old iMac numbers.

For the Mac Pro .... sure that impact will be substantive , but Apple doing a longer refresh ( 2-3 years ) to a smaller crowd is an adjustment. Apple already shifted to an incrementally smaller crowd by increasing the entry price for Mac Pro 100% back in 2019. (and more than 100% if go back to the 2010 era $2,500 entry mark ) . The adjustment already has been established is to iterate slower.

A M4 Ultra with balanced two x16 PCI-e v4 backhaul feeds for the switch and M4 general uplift in CPU, GPU , and NPU would likely do Ok. It isn't going to be an general market workstation 'killer', but it doesn't have to be to pay for itself and keep the line up moving forward.

Going forward it is still going to be case that having more than one internal SSD drive is going to be useful for a subsantive number of folks that do not necessrily need a dGPU. I don't see the Mac Studio really fixing that role ( with 3rd parth xMac chassis to embedded the Studio into).

If Apple doesn't do an "Bigger than Ultra" SoC then there is a decent chance the Mac Pro will get inncrementally smaller ( loose the 1200W power support and a few slots while picking up something like TBv5 while keeping the small SATA options. ). Apple shifting to something that might be a better 2U-3U rack version would make more sense if they are going to 'dump' some major variant of the SoC into their datacenters in larger numbers. ( XCode cloud , AI cloud variants subsets to their datacenter deployments. )

Admittedly this is the same arguments used against the Intel Mac Pro.

The Intel Mac Pro still had the 'Boot Camp' option. Apple is probably going to need to work 'harder' when they don't have that. Either need a 'pass through' VM abilities or need to do some more deliberate work on broadening the PCI-e card market a bit more ( not necessarily opening the dGPU door , but lot less denial about 3rd party SSD market and other accelerator and i/O cards that add value. )
 

deconstruct60

macrumors G5
Mar 10, 2009
12,493
4,053
That, and I hope that increases the chances it includes PCIe5 and Thunderbolt5.

There is little indication that the plain M4 has PCI-e v5 so the bigger die variants probably won't have it either.
Thunderbolt v5 is even more clear. What may see is a DisplayPort 2.1 ( since upgrading the display controller was a priority on this iteration. )

More likely going to see more AI/ML compute from AMX (CPU) , NPU, and GPUs. More dymanic ML upscaling.

It seems relatively clear that Apple was more 'freaked out' about being 'behind' on AI/ML than trying to grab some bragging rights to TBv5 'first to market' win. And they are a couple years behind on DisplayPort v2 versus (Intel AMD) at this point too. That is only going to get more acute in 2025 as more monitors that support it roll out. The dispaly controller was required by the iPad Pro dual OLED and the AI/ML stuff is already firmly entrenched inside the SoC ( so not coupled to PCI-e v5 in any significant way).

If Apple came out at WWDC 2024 and made a big deal about PCI-e v5 AI/ML accelerator cards for MP then maybe v5 would be a priority mid-gen change, but slim changes they are going to do that. Apple's tempid attitude about 3rd party SSD means v5 versions of those also are likely in the same boat. The A/V I/O card focus on the MP 2023 isn't going get a high priority for v5 earlier. Top end WiFi on a x1 PCI-e lane would likely be a bigger push for Apple to 'move up' or same quirk in the design that keeps that at a x8/x16 split in Ultra configurations for the backhaul.
On backhaul lane blockage front, PCI-e v5 is just as likely to bring two x8 PCI-e v5 as it would a x8+x16 v5 one. Apple would take the v5 increase to do a lane allocation reduction to provision equivalent of two x16 PCI-e v4 with less edge space on the dies.


[ for the semicustom SSD module links v5 could buy them fewer lanes, but that has trade-offs. I suspect Apple doesn't feels behind on SSD speeds much right now. M4 isn't pushing that topic so far. ]

I'd expect quasi ECC ( at rest support ) when they move to a new LPDDR version on next upgrade iteration (e.g, M6 ). That may be were PCI-e v5 gets weaved in (if not on M5; to try it out 'in the small' first. ).
 
  • Like
Reactions: ZombiePhysicist

ZombiePhysicist

Suspended
Original poster
May 22, 2014
2,884
2,794
There is little indication that the plain M4 has PCI-e v5 so the bigger die variants probably won't have it either.
Thunderbolt v5 is even more clear. What may see is a DisplayPort 2.1 ( since upgrading the display controller was a priority on this iteration. )

More likely going to see more AI/ML compute from AMX (CPU) , NPU, and GPUs. More dymanic ML upscaling.

It seems relatively clear that Apple was more 'freaked out' about being 'behind' on AI/ML than trying to grab some bragging rights to TBv5 'first to market' win. And they are a couple years behind on DisplayPort v2 versus (Intel AMD) at this point too. That is only going to get more acute in 2025 as more monitors that support it roll out. The dispaly controller was required by the iPad Pro dual OLED and the AI/ML stuff is already firmly entrenched inside the SoC ( so not coupled to PCI-e v5 in any significant way).

If Apple came out at WWDC 2024 and made a big deal about PCI-e v5 AI/ML accelerator cards for MP then maybe v5 would be a priority mid-gen change, but slim changes they are going to do that. Apple's tempid attitude about 3rd party SSD means v5 versions of those also are likely in the same boat. The A/V I/O card focus on the MP 2023 isn't going get a high priority for v5 earlier. Top end WiFi on a x1 PCI-e lane would likely be a bigger push for Apple to 'move up' or same quirk in the design that keeps that at a x8/x16 split in Ultra configurations for the backhaul.
On backhaul lane blockage front, PCI-e v5 is just as likely to bring two x8 PCI-e v5 as it would a x8+x16 v5 one. Apple would take the v5 increase to do a lane allocation reduction to provision equivalent of two x16 PCI-e v4 with less edge space on the dies.


[ for the semicustom SSD module links v5 could buy them fewer lanes, but that has trade-offs. I suspect Apple doesn't feels behind on SSD speeds much right now. M4 isn't pushing that topic so far. ]

I'd expect quasi ECC ( at rest support ) when they move to a new LPDDR version on next upgrade iteration (e.g, M6 ). That may be were PCI-e v5 gets weaved in (if not on M5; to try it out 'in the small' first. ).

Well if it's "just an M4Ultra" I'll probably not get that. ECC is necessary at this point. Without an Extreme version, or at least PCI5/TB5, for me, the 2025 will be pretty much like the 2023, one to skip.

Apple has to stop 'eking' out improvements in the pro sector. It's not what it's about. Then again, apple is making it quite clear, that's not what they are about, and basically HAVE abandoned the pro/enthusiast market. There will be consequences to that. A bridge they will wish they could cross, but have blown up.
 

avro707

macrumors 68020
Dec 13, 2010
2,263
1,654
The Intel Mac Pro still had the 'Boot Camp' option. Apple is probably going to need to work 'harder' when they don't have that.

The solution for the Bootcamp option, as we were robustly told was "to buy a PC" or "build a PC", not by Apple but by folks on social media and forums.

Apple needs to bring back discreet GPU abilities for those people who need to put in some hefty big GPUs to cover the work they need. That won't please the "Apple Silicon or die" folks but it will make the Mac Pro a useful machine for more people rather than a purchase for "cashed up hobbyists". Same with getting Windows compatibility. That will bring it back to the abilities it previously had.
 
  • Like
Reactions: ZombiePhysicist

steve123

macrumors 65816
Aug 26, 2007
1,155
718
Apple has to stop 'eking' out improvements in the pro sector. It's not what it's about. Then again, apple is making it quite clear, that's not what they are about, and basically HAVE abandoned the pro/enthusiast market. There will be consequences to that. A bridge they will wish they could cross, but have blown up.
if Apple was been paying attention there is a glorious opportunity to redefine the MP around AI. A lot of people have been asking for improvements to an architecture that is pretty well baked: PCIe slots, SSD slots, RAM slots. The introduction of AI allows Apple to rethink the MP entirely. Let's hope they show some leadership and make it happen.
 

deconstruct60

macrumors G5
Mar 10, 2009
12,493
4,053
Well if it's "just an M4Ultra" I'll probably not get that. ECC is necessary at this point. Without an Extreme version, or at least PCI5/TB5, for me, the 2025 will be pretty much like the 2023, one to skip.

Mid 2025 is only about 3 years after the MP 2019 got discontinued. Pretty good chance it will still have OS update/security coverage so most of the die hard 2019 folks will likely still be clinging to it. If TBv5 comes with M5 then would catch some folks exiting on year 4 of the countdown clock. M6 if they shifted the big dies to the beginning would catch folks wanting to exit at 5-6 timeframe when things were stopping and Intel support fading faster.

Yeah they probably should have ECC support when get to 256GB , but pretty good chance Apple will hand wave that is mostly "VRAM" and don't 'need it' real bad. (kind of that that initial stab at a supercomputer at VaTech back in the day. )

TBv5 is 'optional' on most of the Intel systems coming in 2024 . ( Lunar Lake only is reported to have TBv4 controllers built in. Ditto for Arrow Lake. So even on Intel side in 2024 only discrete TBv5 controller to get to get that optional throughput (e.g., use the secondary M.2 x4 PCIe-v4 lane allocation. ) Qualcomm and AMD are still largely just avoiding Thunderbolt certifications. Nobody in the industry is rushing to TBv5. It is coming, just not at 'gold rush fever' speeds.

The MBA on third generation M-series is still not TBv4 certified. Those systems haven't even gotten to v4 let alone v5. Apple doesn't have 'gold rush fever' urgency here either.


Apple has to stop 'eking' out improvements in the pro sector. It's not what it's about.

The "Apple has got to deliver future roadmaps " is a very common refrain in this Mac Pro forum for far over a decade. Expectation management matters. The "Rip van Winkle" , 'hobby product' thing is one of the major problems. Get folks on a regular schedule of "buy it or skip it" and things would work better.


Then again, apple is making it quite clear, that's not what they are about, and basically HAVE abandoned the pro/enthusiast market. There will be consequences to that. A bridge they will wish they could cross, but have blown up.

There have walked away from some subsegments of the pro/enhusiast market, but also opened out others. it is is just not the exact same set of customers they are chasing/targeting.
 
  • Like
Reactions: ZombiePhysicist

deconstruct60

macrumors G5
Mar 10, 2009
12,493
4,053
The solution for the Bootcamp option, as we were robustly told was "to buy a PC" or "build a PC", not by Apple but by folks on social media and forums.

Apple's openly stated solution was "virtual machine". The core issue is that their virtual machine tech (that everyone has to use ... via any virtual machine vendor on modern M-series macOS) is not really state of the art. "Pass through PCI-e cards" are supported on a couple of hypervisors.

If Apple doesn't want to do anything at all for a certain class of PCI-e cards then the decent thing to do would be to pass those cards to an OS that did want to do something with them. Apple doesn't have to do everything for everybody, but if don't want to do it... get out of the way.

Apple needs to bring back discreet GPU abilities for those people who need to put in some hefty big GPUs to cover the work they need.


"needs to" ? Apple's Mac business is in a very robust finiancial state. They don't "need to". Similar to how they didn't "need to" sign Nvidia's GPU drivers when they were in a rogue actor state either. Some folks who had a need for specific Nvidia cards might have a requirement , but Apple did not. ( post ban Mac ecosystem units and revenues basically went up. )

What is a "big GPU". VRAM capacity or compute cores? GPUs have multiple dimensions to their attributes.
One set of users is myopically chasing down one vector and another set of customers are looking down another. Neither one of those is strictly an "Apple needs to" if Apple isn't not trying to provide everything for everybody.


That won't please the "Apple Silicon or die" folks but it will make the Mac Pro a useful machine for more people rather than a purchase for "cashed up hobbyists". Same with getting Windows compatibility. That will bring it back to the abilities it previously had.

Apple has dumped UEFI and really has very little interest in persuing the Microsoft Pluton track. It was just a side effect 'accident' that off-the-self Intel hardware also booted BIOS and eventually adopted a more complete EFI implementations. None of that was essential probably of a Mac as prior to x86 transition Apple didn't touch BIOS with a ten foot pole either. The whole point here is to not build off-the-shelf SoCs. Primariy objective is to make a Mac SoC not random OS SoC. Apple doesn't sell just the SoC at retail , they sell whole systems.
 

avro707

macrumors 68020
Dec 13, 2010
2,263
1,654
Apple has to stop 'eking' out improvements in the pro sector. It's not what it's about. Then again, apple is making it quite clear, that's not what they are about, and basically HAVE abandoned the pro/enthusiast market. There will be consequences to that. A bridge they will wish they could cross, but have blown up.

We can make our cases clear to Apple directly rather than going through long-winded forum debates and if Apple doesn't care then we go elsewhere when our current machines are no good.

That’s what will end up happening anyway. My 7,1 is will my last Mac Pro unless I snap up another 7,1 from somewhere to replace one of the 5,1 machines.
 
Last edited:
  • Like
Reactions: ZombiePhysicist

edanuff

macrumors 6502a
Oct 30, 2008
578
259
I expect to see the next Pro and Studio models to be positioned as fairly strong Gen AI machines. Developers are seeing fairly remarkable price performance on these already for running LLMs and more unified memory and storage options would put them in a very good spot in the benchmarks.
 

impulse462

macrumors 68020
Jun 3, 2009
2,097
2,878
I expect to see the next Pro and Studio models to be positioned as fairly strong Gen AI machines. Developers are seeing fairly remarkable price performance on these already for running LLMs and more unified memory and storage options would put them in a very good spot in the benchmarks.
If they support training models with the ANE yes. As it is right now, even running existing models at inference time doesn’t use the ANE at all, although performance is pretty decent.
 

deconstruct60

macrumors G5
Mar 10, 2009
12,493
4,053
If they support training models with the ANE yes. As it is right now, even running existing models at inference time doesn’t use the ANE at all, although performance is pretty decent.

Apple's ANE/NPU is likely geared toward inference , not training. If it is skewed toward INT8 or BF16 then may not train at FPT6 all that well. Pretty good chance that high "TOPS" number they like to quote is INT8 or smaller. For wider formats the GPU may actually have more FLOPS than the NPU ( on a Pro/Max/Ultra ).


Also pretty good chance the NPU subsysystem (like the CPU clusters) doesn't have access to the same max bandwidth that the aggregate GPU has. Training dataflows are different than inference.
 

deconstruct60

macrumors G5
Mar 10, 2009
12,493
4,053
WWDC 2024
  • M4 Mac mini
  • M4 Pro Mac mini
  • M4 Max Mac Studio
  • M4 Ultra Mac Studio
  • M4 Ultra Mac Pro


I wouldn't bet on that. Rumors point toward Apple putting both 'regular' and 'Pro' iPhone on A18 (and A18 Pro) . If both of those are also moving up to N3E , TSMC may be capacity constrained for much of June/July just with the iPad Pro and the ramp for the phones (that is about twice as big as the last couple of years where the Plain/Pro were splitting fab processes. )

Throw on top Intel ramping on N3B. ( means primary iPhone SoC getting 'off of' N3B is somewhat going to be backfilled with Intel. Which means TSMC cannot switch EUV fab machines over to N3E. ) ASML only makes so many and now Intel, Samsung, and TSMC are buying them. By end of 2024 other folks will be ramping on N3E .

If Apple is skipping M3 Pro/Max for the Mini/Studio then there likely will be a hiccup; not an increase in rollout speed.
 

impulse462

macrumors 68020
Jun 3, 2009
2,097
2,878
Now that apple intelligence has been announced, i'm curious if this changes the thought process for people in this thread for the next mac pro
 

deconstruct60

macrumors G5
Mar 10, 2009
12,493
4,053
Now that apple intelligence has been announced, i'm curious if this changes the thought process for people in this thread for the next mac pro

The folks pointing at "Apple is going to do a dGPU because they have some silicon that runs a lightweight iOS fork". It is't a GPU. It is a specialized appliance server (that only does inference) and a thinned out iOS is the "whole OS" the system runs. ( no persistent local user storage at all. Boots iBoot for the thinned out iOS. ).

Apple is likely going to get enough volume deploying these as specialized inference only servers to keep the Ultra SoC going like it is (with evolutionary improvements) . It isn't a modular, ML training GPU killer. The next Mac Pro is probably going to be structured like the 2023 MP.
 
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.