Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
Are you saying someone who needs a Mac Pro is going to ask around, "well, gee, I don't really know what RAM is; how much do I need?"

If they don't know, they definitely don't need a Mac Pro.

Not sure what you mean here.

Why can't your brother get a Mac mini and any ol' display?
He just may have to do that. Doesn't change the fact he would like to get an AS version of what he already has. I know he's not alone. Why can't Apple offer it?
 
Finally, Apple comes to its senses, scrapping the overpriced, overweight Pro for both the Mac Studio AND an M5 Ultra chip- woo hoo!

Next, dump the Vision Pro until it's the size of a Walkman headset.

Best as always,
Loren
 
The real problem is the people that need the bandwidth know that the Mac Pro PCIe is a lie. The processor has 24 lanes, 8 of which are internal to the chip. That leaves 16 lanes shared between Thunderbolt and 7 PCIe slots. If you were to put a single x16 card in it could use all of the bandwidth available on the processor, so no, the PCIe isn‘t actually better than Thunderbolt it just looks like it could be.

The Xeon can run all those slots at full chat natively, having over 100 lanes.
But the M2 Ultra has 32 lanes (same as the Power Mac G5 and first Mac Pro) with 8 dedicated to the internal SSD. The other 24 lanes are available to the PCIe slots which actually can use up to 64 lanes but have to share those 24 lanes. So it is like the older Mac Pro's and Power Mac's of yesteryear except PCIe 4 speeds instead of v1 speeds.

I think this is why they were planning and betting on the Extreme chip for the Mac Pro which would have this all taken care of for sure.

With the M2, if you run one card in x16, you can have full PCIe 4 speeds and still have 8 left over for other less demanding cards. Not ideal but a lot better then what you were saying above of having zero left over. Edit: Also, if a card is not currently in use, it uses zero lanes. With the Apple Card in, it appears 88% is used up.

I would be hoping that the M5 Ultra will solve this problem, just like future Mac Pro's improved and eventually solved that situation over time. Edit: But the Mac Pro 2019 also only had 64 lanes and had more x16 slots available so also had the same pool situation just more lanes were available overall and there was nothing more than that at the time.
Mac Pro 2023 PCI info: https://support.apple.com/en-us/104947

Also, another resource to explain the PCIe compared to 2019: https://softron.zendesk.com/hc/en-u...-Ultra-Limits-and-possibilities-of-PCIe-slots
 
Last edited:
  • Like
Reactions: darkblu
Then by 2016, a manager convinced Ternus to start a new Mac Pro, and out came the 2019, with Afterburner, MPX and all. But meanwhile, a different team was already working on the M1, making much of that work moot.

the Mac Pro 2019 was in some respects an experimental lab for several things Apple wanted to work on for Apple Silicon.

1. GPUs hooked together with a high speed bus. Vega DUO , W6800 DUO. two W6800 with a bridge. I don't think Apple was looking for a 'distant' bridge for Apple Silicon, but just getting more graphics library suppose in case there were more NUMA artifacts in the solution they chose. ( an 4 * Max was going to be far more likely to have those than the 'DUO' Ultra would).

In short, wanted apps to spread out over more cores relatively somewhat seamlously .

2. Afterburner in retrospect is a 'public trial' of their ProRes decoding logic. FPGA so easy to fix as went to market. (also to possible tap dance around legal 'drama' if that popped up . ). Make it more mature before folding it into the die of the whole line up. The 'plain' M1 didn't get ProRes. That likely was in flux for a while (in addition to consuming more space. ).

3. modest PCI-e Backhaul fan-out through a large 'sever' PCI-e switch. The Ultra was very likely going to have limited lanes ( smaller than the Xeon W class ).

4. Mostly unified graphics and Thunderbolt. ( a step toward integrated TB. ). Nominally,
Somewhat less 'experiment' but somewhat new to this particular product. MPX wasn't 'moot' it was a test harness to try out the concept.

Similarly the "Apple I/O" card which also evolve into the Apple Thunderbolt PHYS card (once finished integrating
controllers in the SoC).


5. T-series. ( like the rest of the mac line up. The iMac was very oddly very late to this, but 'T' was pretty obviously for 'Transition'. The Xeon W series lacking QuickSync (since no iGPU) would make the Mac Pro an odd-duck where MBP 15" would slice and dice standard video formats faster. ( Apple had issues with AMD video en/decode ) .

Again Intel QuickSync was replaced largely in full by Apple's own implementation in Apple Silicon ( Apple dragged their feet on AV1 support , H.264/H.265 integrated. )


The real likely stumbling block that illuminated that the iMac Pro had some of same limitations as the Mac Pro 2013 would the trajectory that GPUs had of extremely power hungry. AMD/Nvidia were in a pissing match where power consumption didn't matter. Intel was also fumbling where the workstation roadmap was for "even hotter" chips of hiccuping fab process. The iMac Pro largely had the same max power envelope limitations that the MP 2013 had.

The iMac Pro was hobbled in part for updates because Intel didn't have anything substantively better for the thermal constraints in 2018-2020 anyway.

Intel and AMD were doing lots of stuff that Apple though was 'wrong' in terms of efficiency. So Mac Pro 2019 is more a stop gap until they could step off. If AMD got their act together a bit Apple could add some more 'stop gap' MPX updates. If Intel bungled until the end they could coast on the 2018-19 offering.


Because are sales are so low that it isn't worth it.

It isn't just low volume. It is also higher cost for the major components. And no other products to put them in. The plain 'Mn' processor gets a 'hand me down' to the iPad Air and also fans out into the iPad Pro. The M5 Pro is in the MBP 14/16" and Mini "Pro". The Max is reused in the Ultra (so far). The Ultra has what "hand me down" ? Not really. Some are hand waving at Private Cloud Compute (PCC). That seems likely to be temporary. If the Apple made the Mac Pro get a 'hand me down' Ultra 3 when the Studio get a M5 Ultra .

The Ultra SoC is too expensive to develop only to throw it into the trash can every 12 months. Even if Mac Pro unit volumes were 10x higher that likely still would be the case. Apple doesn't throw throw the 'Max' away every 12 months because the Studio and Mac Pro are committed to consume them for longer.
 
  • Like
Reactions: MacHeritage
I would be hoping that the M5 Ultra will solve this problem, just like future Mac Pro's improved and eventually solved that situation over time. Edit: But the Mac Pro 2019 also only had 64 lanes and had more x16 slots available so also had the same pool situation just more lanes were available overall and there was nothing more than that at the time.
Mac Pro 2023 PCI info: https://support.apple.com/en-us/104947

As long as the Ultra is composed of two "Max" dies it likely would not result in a lane count expansion. Apple might move to 24 PCI-e v5 lanes. That still give the ability to assign one x16 bandwidth assignment through the PCI-e switch to some slot, but leaves a bigger budget for 'the rest' to share. That really didn't completely solve the problem from the purest viewpoint of direct point-to-point connect to SoC for a single x16 slot.

Apple is probably going to so with a 'fan out' after get off the die. Apple's "poor man's HBM" approach to memory controllers demands lots of lanes to provision larger memory bandwidth. That is extremely likely is a higher priority than PCI-e lane count bragging rights.

Is Apple going to go the PCI-e v5 just for the Mac Pro ( the PCI-e lane provisioning for the Studio ultra is moot. 'road to no where' in that system. If they are not super motivated by numbers, then I wouldn't hold my breath. It would likely help , but some commentary from Apple is about looking backwards in time at the PCI-e cards they are targeting, not forward.
 
  • Like
Reactions: MacHeritage
The real problem is the people that need the bandwidth know that the Mac Pro PCIe is a lie. The processor has 24 lanes, 8 of which are internal to the chip. That leaves 16 lanes shared between Thunderbolt and 7 PCIe slots. If you were to put a single x16 card in it could use all of the bandwidth available on the processor, so no, the PCIe isn‘t actually better than Thunderbolt it just looks like it could be.

The Xeon can run all those slots at full chat natively, having over 100 lanes.

There is lots of propaganda about the PCI-e provisioning in the Mac Pro 2023.

I'll start with the Xeon comment and then more forward for context.

The 2019 Intel option for a Xeon W CPU only had 64 lanes. (in some years prior to 2019 it was less)


The 2021 Intel option for a Xeon W CPU only had 64 lanes.

in 2023 the W5 only had 64 (yes the W7 and W9 had more, but that is a higher minimal system cost. This W5 ~ $800 list )

in 2024 the W5 has 112 but the price has jumped to $1699. ( same microarchitecture family just upclocked and bigger to try to be more competitive with AMD).

Apple announced they were going to dump Intel in 2020. So this > 100 lane solution didn't exist until 3-4 years after Apple said they were leaving. Whose is dealing in deception here? macOS on Intel goes largely comatose in 2026 (It has been anounced). If Apple had introduced a new Intel Mac Pro in 2023 and then said updates were dead in 2026 that would be closer to 'a lie' that it was a viable , long term platform for support.

Every Intel Mac Pro has a PCI-e switch in it. 2006-2019. That the 2023 Mac Pro has a PCI-e switch in it is much more continuity than a 'lie'. The Mac Pro 2019 introduced the "Expansion Slot Utility" app which allowed user control of how the server grade, dual input PCI-e switch allocated bandwidth.

The change with the Mac Pro 2023 was , relative to MP 2019, Apple tossed out MPX slot 1 and MPX slot 3.
The backhaul to the server class PCI-e switch went from x32 PCI-e v3 to x24 PCI-e v4 ( equivalent of x48 PCI-e v3).


The PCI-e for the Thunderbolt is embedded in the TB controller. There are no discrete TB controllers in the M-series. TB controllers are integrated into the die and that PCI-e root is on the internal bus that function units communicate with. Similar to what Intel does with its die embedded TB controllers. It is far more energy efficient.

As other post noted the PCI-e 'diversion' is for special case use for SSD controller to NAND communications.

Which brings to the Mac Pro 2023 'lie'.

The Mac Pro 2023 has 6 slots ( no counting the half Apple I/O slot).
two x16 slots
four x8 slots
I/O slot ( half x4)

The Mac Pro 2019 had 7 slots ( no counting the half length Apple I/O Slot).
[ collapsing the MPX bays into just slot count ]
four x16 ( 1 from MPX bay 1 , 2 from MPX bay 2 , 1 above bay 2)
three x8 ( 1 from MPX bay 1 , 2 above bay 2 )
I/O slot ( half x4 ) four x16 slots


So the Mac Pro 2023 has fewer slots.

Pragmatically Apple dumped the two 'primary' x16 MPX slots. (along with the additional TB provisions that loaded down that PCI-e switch). the CPU to GPU bandwidth on die in M-series is a major uplift from from PCI-e 3,4 ( or 5,6). The thinning out of bandwidth for TB disappeared also.

With 'old' Slot 1 and Slot 3 dropped, it is a net uplift. "But I could put in a GPU card to drive the GUI in Slot 1 or 3 " ... on macOS on Apple-Silicon there are no GPU drivers for 3rd party GPUs. Apple never promised they were going to make one. (in WWDC 2020 was totally empty on the matrix . This is before shipping any M-series Macs. ). If there is no software to run a video card in a slot, why would Apple provision a video card slot? The 'lie' there is that there is an option to it when it doesn't work. Which Apple did not do.

The 'lie' is in the denial that 'old' Slot 1 and 3 were not removed. Apple never said there were there. Coaching that as a 'lie' is propaganda at best. The count of x16 slots went down two in the specs and discussion of the system.
 
Is Apple going to go the PCI-e v5 just for the Mac Pro ( the PCI-e lane provisioning for the Studio ultra is moot. 'road to no where' in that system. If they are not super motivated by numbers, then I wouldn't hold my breath. It would likely help , but some commentary from Apple is about looking backwards in time at the PCI-e cards they are targeting, not forward.
This is why, if they are designing the new Ultra chips separately from the Max chips, this might improve (I am holding out some hope!) but all we have are some rumours that they are doing that. The only confirmed parts we have is the new chiplet CPU tech coming to the M5 Pro, Max and Ultra and that is why the delays for the M5 Pro and Max.

All around great posts above! Greatly appreciate the details.
 
It’s called a Windows PC. If you can handle Linux, you have the expertise to set up a local account and disable telemetry (one click with O&O Shutup), or whatever other reason makes Windows ‘impossible’ to use.
No, Windows is a **** ass ugly OS with weird design choices everywhere lol. My favorite thing is how jerky and laggy the UI feels compared to any recent version of gnome/kde or macos.

But yea, for the record, my PC is setup to boot into fedora 43 by default. I have windows 11 on there, and yes I did set it up with a local account and disable telemetry! Fun times.... still doesn't help that I think windows is a **** ass ugly OS lol.

Like, spend $2000 on a second machine to handle commercial photo/video/music software that I want to use, but I can't run from Linux? ABSOLUTELY, YES PLEASE! THANK YOU. I've been happy enough with my Mac Studio.

Ultimately doing that is still cheaper than buying a mac pro lol.
 
The 2019 Intel option for a Xeon W CPU only had 64 lanes. (in some years prior to 2019 it was less)
Yes, older chips were less capable than current chips. If that surprises you then you’re on the wrong forum.

Xeon PCIe count is per socket so that’s still 128 lanes in a dual socket system. No matter how you dress it up, the PCI used to be more capable for expansion and current design uses shared bandwidth, which means devices will throttle badly.
 
Pragmatically Apple dumped the two 'primary' x16 MPX slots. (along with the additional TB provisions that loaded down that PCI-e switch). the CPU to GPU bandwidth on die in M-series is a major uplift from from PCI-e 3,4 ( or 5,6). The thinning out of bandwidth for TB disappeared also.
On-chip BW is only one side of what Apple are doing with their SoCs. The other is called 'memory coherency' and is what the APU paradigm offers. PCIe just does not jive there, and that is one of the reasons for the existence of e.g. NVLink. Yes, CXL does coherency on top of PCIe5, but that's a pie-in-the-sky compared to what Apple et al are doing.
 
All five of them? Is there anyone on this website who owns an AS Mac Pro?
FWIW about a half dozen of the folks under my bosses umbrella (of 25 direct reports) use them. It comes up every so often because I’m jealous in theory, but in practice I personally have no need of one - any major heavy lifting I need that the MBP cant provide is going to happen either on much bigger on prem hardware in our DC or on the even bigger cloud footprint I work on. And my laptop is mobile. If I had an actual business need for it the company would buy it, it’s a rounding error on the cost of an engineer on our team (hell I personally, not shared team resources or our actual production environments, just for my test environments, spend as much of company money on AWS *every month* as that machine costs) - but I don’t need one, I just kinda want one lol.

Tldr it’s a product aimed at companies, there are folks who use them, and for the folks that need them companies will shell out for it because unlike it being a big purchase for a consumer it’s a tiny amount in the cost of a developer or other employee that needs it for work
 
Last edited:
Tldr it’s a product aimed at companies, there are folks who use them, and for the folks that need them companies will shell out for it because unlike it being a big purchase for a consumer it’s a tiny amount in the cost of a developer or other employee that needs it for work
Very sensible post and quite right. I think the angst comes from the very narrow band of folk it does fit. Slightly more capable than the studio but not so much so that you need a lot of bandwidth is a really specific niche. The Studio is so very capable for most requirements now that that band is shrinking, especially with no CPU update on the pro. We’re pretty much now at the stage where the Studio can handle everything and those that need more the Pro doesn’t really help.
 
FWIW about a half dozen of the folks under my bosses umbrella (of 25 direct reports) use them. It comes up every so often because I’m jealous in theory, but in practice I personally have no need of one - any major heavy lifting I need that the MBP cant provide is going to happen either on much bigger on prem hardware in our DC or on the even bigger cloud footprint I work on. And my laptop is mobile. If I had an actual business need for it the company would buy it, it’s a rounding error on the cost of an engineer on our team (hell I personally, not shared team resources or our actual production environments, just for my test environments, spend as much of company money on AWS *every month* as that machine costs) - but I don’t need one, I just kinda want one lol.

Tldr it’s a product aimed at companies, there are folks who use them, and for the folks that need them companies will shell out for it because unlike it being a big purchase for a consumer it’s a tiny amount in the cost of a developer or other employee that needs it for work
Curious: What do they use them for?
 


Apple's high-end Mac Pro desktop computer is currently "on the back burner," according to the latest word from Bloomberg's Mark Gurman.

Mac-Pro-Feature-Blue.jpg

In his Power On newsletter today, Gurman said he heard that Apple has "largely written off" the Mac Pro, with the sentiment inside the company being that the Mac Studio represents the present and future of Apple's pro desktop computing.

Apple is working on a high-end M5 Ultra chip, but Gurman said the company is currently "only" focused on a new Mac Studio with that chip, which leads him to believe that the Mac Pro "won't be updated in 2026 in a significant way."

It sounds like Gurman has not heard anything about a Mac Pro with an M5 Ultra chip being in the works, but he has not entirely ruled out the possibility.

In recent years, the Mac Pro has become redundant for many customers, as it is essentially a larger and more expensive Mac Studio with PCIe expansion slots. Apple last updated the Mac Pro with the M2 Ultra chip in June 2023, and Gurman's wording suggests that the future of the computer may be in jeopardy.



Article Link: Mac Pro Reportedly on 'Back Burner' and 'Largely Written Off' at Apple
As someone with a 2019 Mac Pro, I’m surprised by the extensive disregard for the Mac Pro in the comments for this article. I value the flexibility of being to upgrade the Ram and gradually add internal storage. I would love the option of a more up-to-date one since the latest operating system is not playing well with the Intel chips in my 2019 model and I wouldn’t want to go to an M2 with newer versions of the M chip available for other Mac’s. A gaggle of external wires for peripheral storage awaiting me if I have to go to a different type of newer Mac is not a prospect I relish.
 
Curious: What do they use them for?
The folks I was talking about specifically do QA work, they do a lot of large scale local testing and write a lot of automated tests

There are plenty of other folks in the company that use MPs too though, lots of design etc folks, some devs, etc.
 
Last edited:
. A gaggle of external wires for peripheral storage awaiting me if I have to go to a different type of newer Mac is not a prospect I relish.

In many cases (not all but enough to make a difference), it is not particularly necessary to have "external" cables if spend money on a container. For example:


Mini shrank enough can put optionally put two in addition to cards.

For example, the xStudio/Echo III has three PCI-e slots. Could start off with one , then two SSD carry cards and got from 1-2 SSD to 4-8 SSDs relatively easier for less than the Mac Pro ( $3,999 + 1,799 = $5,699 ) . If didn't 'need' an Ultra SoC then even less ( full Max , 128GB RAM , 0.5TB SSD $3,499 and closer to an even $5K price. )

Similar examples of a Mini Pro coupled with one or two card extenders in a single rack where don't need Max (or better) 'grunt' from the SoC.

Folks using a J2 sled with two SATA HDDs inside a Mac Pro 2019 had 'gaggle of wires' also... they just were inside the container.

There is little chance Apple is interested in making something that is primarily just a container for commodities. They really were not in the Mac Pro Intel era either. There just happened to be some overlap there driven by side effects of parts Apple used not under their control.

As someone with a 2019 Mac Pro, I’m surprised by the extensive disregard for the Mac Pro in the comments for this article. I value the flexibility of being to upgrade the Ram and gradually add internal storage. I would love the option of a more up-to-date one since the latest operating system is not playing well with the Intel chips in my 2019 model and I wouldn’t want to go to an M2 with newer versions of the M chip available for other Mac’s.

It is not a 'not playing well' issue. It is more so that 10's of millions of folks moved on from paying for macOS on Intel. Very similar with the old workstation CPU in the Mac Pro . The Xeon W are just binned on clock and reconfigured Xeon SP processors. Pragmatically the server customers paid for those and getting workstation product is a happy side effect. If the 'sugar daddy' customers all go away, then it really isn't a 'workstation value' issue. The need to back fill all that missing money typically passed up the "I'll pay more money because it is valuable" threshold for most folks. The workstation folks were not willing to pay the whole freight before and not likely to want to pay the whole freight now either. Humans typically price anchor on costs and it is typically very hard to get them off. ( much of what goes into Apple's. 'xx99' pricing. Like it is always a whole dollar short on bill of material costs; not. )


Apple isn't selling individually, hand crafted Mac Pro to people. They are selling into a market. So it is the aggregate value perceptions and choices made that matter. Some of the grufty comments in the thread contributes to why Apple is disinterested. ( there is always a faction of "want 2012 pricing for a container box" faction in most of these Mac Pro threads. ) Customers who buy at a slower refresh rate also are likely contribute to disinterest ( "Apple sell me a container and then for more than several years I will buy from other folks besides you". Well that kind of makes them a non customer. )
 
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.