Does anyone expect one or more PCI Express 6.0 (or even 5.0) x16 slot(s) on the next Mac Pro?
PCI-e v6 isn't coming for a couple of years even from the folks deep in the server SoC business. It is unlikely Apple would adopt that any time soon. It just past ratification this year and probably not going to make 2023 updates from almost anyone since design cycles last a couple of years. 2024 is likely earliest real window for that.
PCI-e v6 would be disconnected from Apple's general approach so far with the M-series work. Apple has mainly taking PCI-e v4 as an opportunity to provision "high" bandwidth with fewer lanes. It hasn't been to greatly increase higher overall aggregate bandwidth or to enable heterogeneous implementations of accelerators. They have taken die edge space that might have gone to PCI-e and thrown it at more memory controllers and associated memory bandwidth and , on the M1 Max, to UltraFusion. Relatively large number of PCI-e lane allocation has not been a priority. More to "System on a chip" than to "multiple , different chips".
So careful for what wish for. 64 PCI-e v3 lanes could easily turn into 32 PCI-e v4 lanes with Apple's approach. But if greatly pound the table in some high version number feature checkmark 'war' ... could end up with just x16 PCI-e v5 lanes and x8 PCI-e v6.
PCI-e v6 without CXL 2.0(or better) is likely going to be less than effective. v5 and v6 are coming with distance overhead (requiring repeaters). Apple lack of participation in CXL makes it seem unlikely that there are trying to move aggressively with v6 or v5. ( And harder to do those at high Perf/W ratios now. ) Pretty good chance Apple takes a "not invented here" attitude toward CXL and does nothing for at least a couple generations ( if not much longer). Probably not going to be a widespread variety of commodity PCI-e v6 devices . v4 , v5 , v6 is a progression into deeper niches. Doubtful Apple is going to rush full speed into that narrowing. They may be a tail end adopter to pick up which use cases get very high traction once they mature , but not liking to take the hiccups of early edge adoption.
Where Apple is at is four x1 PCI-e v4 lanes. Just getting them to the point they provision two x16 bundles would be a big leap for them (without any version change). Let alone jumping far ahead of the curve on PCI-e version adoption. The other issue is that v5 or v6 at high lane provisions would be more bandwidth pressure on their internal bus network. That probably isn't as high a priority of keeping the GPU cores feed and expose limitations of UltraFusion super-wide and lower clocked connector.
Two x16 PCI-e v4 bundles fed into an upgraded version of the PLEX switch present in the Mac Pro 2019 could provision six PCI-e slots. ( basically removing slot 1 and slot 3 from MP 2019's set up). Pretty good chance Apple would present that as a sideways (no backsliding) move from the MP 2019 ( even more so if didn't support 3rd party GPU cards.. which are primary targets for 'old' slot 1 and 3. )
IMHO, Apple might pick up PCI-e v5 around M4-M5 generations. Discrete Wi-Fi 7/8 chips might push them along. The second generation Mac Pro would feel more 'heat' trailing far behind. Small chance v6 could be never ( if Apple doesn't let compute GPGPU/accelerators into the mix over time.) .
Recent history on PCIE slots
The 2019 Mac Pro has two PCI Express 3.0 x16 slots, each capable of 15.39 GB/sec. (bidrectional), just a hair less than 25% of the bandwidth the latest PCIE 5.0 x16 slots found on the latest PC motherboards that started shipping last month. ('19 MP also has a bevy of x8 slots and an x4 slot that comes occupied, but this post is concerning support for GPUs that require x16 slots.)
The MP 2019 has four x16 electrical slots ( 1 , 3, 4 , and 5). Slots 1 and 3 are directly connected to the CPU. The majority of the slots of the MP 2019 are not directly connected. Most of the slots are devoted to distributing bandwidth out of a higher number of add-in-cards rather than deliver maximum bandwidth to just one card. The current Mac Pro put about as much R&D focus on slots 4 and 5 as on 1 and 3.
Similar on the MP 2009-2012. The two x4 slots are hanging off a PCI-e switch. The 2019 has better backhaul, but similar set up.
Slot 1 on a Mac Pro from 2006 on has primary been aimed at a boot screen GPU. Apple has gone in a different direction on boot screen GPU with the M-series.
Myopically focusing just on slots 1 and 3 missing the 'forest' for a single 'tree'.
Recent history of Thunderbolt & OCuLink ports
Meanwhile, Thunderbolt 4 ports -- the best found on any Mac -- each provide a paltry data rate of 5 GB/sec (40 gigabits/sec, bidirectional -- equivalent to a PCIE 3.0 x4 slot).
Thunderbolt 5 (just previewed last month by Intel, not released yet) is not going to rectify this situation. All it does is to double the bandwidth of Thunderbolt 4 (which BTW is the same as Thunderbolt 3), and there's no release date.
PCI-SIG working group (as part of the PCIE 4.0 spec) updated another kind of port, OCuLink, to version 2 (OCuLink-2),
OCulink is just misdirection. Apple isn't going there. OCuLink has substantively even weaker implementors uptick than Thunderbolt does.
Thunderbolt '5' triples the bandwidth of TBv3 in specific , but popular corner cases. 120 Gb/s outbound and 40 Gb/s inbound is an option from 80Gb/s bi-directional. That will make multiple display video out (or super larger and fast resolution to single screen) more viable. Probably would not be useful for eGPUs (which really isn't the focus). The PCI-e provisioning 'backhaul' to Thunderbolt likely isn't going to change with "v5". There may be an eventual iteration substantively further down stream but probably not coming 'soon' or intermediate term.
Without 3rd party display GPU drivers, internal or external PCI-e slots doesn't really make a lick of real world difference.
Thunderbolt '5' is hooked to USB4 version 2. The catch-22 for TB at this point is that it is tightly coupled to USB. USB moves slowly. It is a very broad community run by a deliberately broad committee. It isn't just one-two companies pushing the 'standard' onto everyone else. Have to get broader buy-in which typically takes much more time. AMD is still not at the stage building TB protocol controllers inside their SoCs so still have ways to go on adoption breadth. ( DisplayPort 2.1 means have somewhat cracked their GPU packages so incremental process. )
Thunderbolt has failed to live up to its promise as a replacement for PCIE slots
Thunderbolt's original promise was to replace the need for PCIE slots, but clearly it will never meet that promise in the foreseeable future. So why does Apple still release desktops where the only expansion is Thunderbolt?
This really was the primary point of Thunderbolt. A combo of PCMCIA/ExpressCard and USB utility? Yes. One plug that does "everything" ... perhaps. ( Intially tried to subsume next-gen USB socket but went to mini-DisplayPort when USB-IF brushed them off. After had substantial traction and USB-IF at port inflection point (Type-C) finally 'merged into' USB-IF with much less resistance. ) But a 100% complete replacement of every single workstation Add-in-card use case.... no.
Why does Apple release desktops with only Thunderbolt? Mainly because they mostly want to make laptops and mobile. It is vast bulk what they sell. The most subcomponent parts in common gives Apple better economies of scale of paying more affordable prices (cost down , margins up). The other major contributor is that they are selling whole-system solutions; not DIY cobble your own system solutions. Most folks want to buy a system that works out of the delivery box ( unpack , plug it in, works for more than several years.). The objective is to use it; not play/tinker inside of it. Nor the objective to have someone else put put other stuff inside and Value-add-resell to end users.
As you are all aware, in 2013 Apple launched a new Mac Pro without PCIE slots, based on the idea that a host of Thunderbolt 2 ports and a couple of proprietary graphics card slots would be good enough to satisfy pro users' needs for expandability. It was an abject failure -- nobody ever made another GPU for those special slots (correct me if I'm wrong!), Thunderbolt 2 enclosures with PCIE slots were severely bottlenecked from offering the bandwidth required by GPUs, and pro users largely rejected and ridiculed it.
eGPU has much bigger problem with lack of macOS driver support than hardware limitations. macOS was a late adopter of eGPU support. macOS on M-series dropped that support. To a very large extent this is not a hardware issue.
Apple didn't even make another GPU for those special slots. Pointing at lack of 3rd party hardware updates is more than bit of misdirection. With no Apple approved GPU drivers there couldn't be any other solutions.
There was multiple dimensions to the problem there. If AMD had an appropriate upgrade solution Apple might have used one. They didn't deliver either. ( in that time frame AMD delivered Fury (which was too hot and expensive ) and Polaris ( which was only a sideways performance move. In all fairness, AMD was mainly trying to not go bankrupt at the time. Keeping the lights on was a challenge. ). Nvidia was in the process of blowing up their relationship from Apple. So there wasn't a viable solution anyway. Thunderbolt had exceedingly little to do with that.
The Mac Pro 2013 two GPUs basically played the role as the slot 1 and 2 x16 played in the MP 2009-2012 did. Thunderbolt in MP 2013 was only really covering the 'old' x4 slot 3 and 4 bandwidth roles in the older systems. Apple directly directly supported eGPU role with TBv2 at all. ( folks did hackery to get MP 2013's to drive eGPUs. Apple wasn't actively promoting that. Official Apple support came with TBv3 systems. Intel didn't make eGPU support a formal (optional?) requirement for Thunderbolt until TBv3. Again pointing wasn't being pitched as a 100% PCI-e standard card format replacement if GPU class not even in the certification requirements. ).
Many of us cMP owners waited and waited and waited for Apple to come to its senses. Three years ago, Apple finally gave us the 2019 Mac Pro, seeming to signal they had heard the complaints. However, the 2019 Mac Pro was priced out-of-reach for most of us who had waited so many years expecting something in the $3500-5000 range.
Apple went for a very high margin product that counterbalanced the low volume market it would sell into. They really were not primarily going after enabling the cheapest commodity add in components. By 2017 they likely knew there were not primarily shooting at discrete GPUs in the long term.
The "leaned too hard on Thunderbolt" was acknowledged in the MP 2019, but they didn't walk away from TB at all. In fact, substantively custom infrastructure was to enable TB provisioning ( to top/back TB ports ). Got a rack enclosure that could pack lots of cards inside the container. But 4-5 HDX Avid cards isn't the affordable commodity market focus.
They did a substantial provisioning shift from was the "one and only one internal drive" , but not for direct from Apple Build-to-Order configurations. ( add your own internal drive sleds compromise).
Then Apple announced they were ditching Intel entirely, followed by not issuing a 2020 or 2021 version of the Mac Pro featuring the latest Xeon processors, yet still charging the same price for the 2019 Mac Pro as when it launched in 2019, making it exceptionally difficult to justify purchasing if (like me) you finally saved up enough scratch to afford it.
Ditching just Intel? Apple ditched Nvidia before they ditched Intel. That chasm of an effective MP GPU hole that AMD left in the 2014-2017 time frame also didn't help much with long standing commitment to them either.
There are over 50 supported PCI-e add-in cards for macOS on M-series (Arm) . Apple didn't have major delivery and/or driver dust ups with those folks. And surprise there is support for them. (although some still not on DriverKit and still coasting on kext (kernel extensions which are deprecated ).
Waiting on what? From PowerMac -> MacPro 2019 the entry prices for "Mac Pro class" systems have going from about the $1,799 zone , to $2,199 zone to $2,599 zone to $2,999 zone to $5,999 zone. At no point there did Apple backtrack and go significantly backwards on system price in this Mac Product classification.
Waiting for some $2,199-3,999 slot box to magically reappear? Probably not going to happen.
Yet, three years later, there is still nothing,
2009 - 2012 ... three years and essentially nothing 'new' then either (in terms of base motherboard and core platform) So what is exactly 'new' here with this 3 year gap? It has been over a decade since Apple did fast iterations on the whole Mac Pro platform.
If go glance at the Dell/HP ( Dell 5820 , 7820 , 7920 . HP Z6 G4 , Z8 G4 ) there is a a bunch of Intel SP gen 2 stuff there still for sale also. ( the Xeon 6300 was largely a 'bust' at the major workstation vendors. There is some relatively recent uptick with Threadripper options in 2022 , but the major players 'sat' on the Intel Gen 2 stuff. Still are. ). So really not just Apple. If Intel hadn't choked so hard on W-6300 in terms of power consumption and lateness for volume availability... there might have been a Mac Pro 2021 with W-6300. But they didn't. It isn't 100% on Apple. They have contributions to nothing moving here , but not 100% on just their efforts.
other than a Mac Mini-shaped trash-can do-over with all the same lack of expansion and reliance on woeful Thunderbolt for everything.
The Mac Studio took the iMac 27" place in the line up. There really isn't a 'change' there at all in reliance on Thunderbolt. It does make it so that some more folks that need 2006-2012 Mac Pro performance levels don't have to get herded into buying a Mac Pro as the only option. Same progress-over-time thing that has shifted overall PC sales from desktops to laptops. Fewer folks "need" a desktop to get enough performance.
Mac Studio just make relatively old Mac Pro user base smaller. The Mac Studio still has backorder issues over six months after introduction ( BTO Ultra configurations are "into December" lead times). You can throw derogatory "trash can" at the Studio but it is selling in substantive (and highly like profitable) numbers. It looks nothing like a trash can at all either literally or figuratively.
Will Apple ever come to its senses, or should I just switch to PC?
During the "abject failure" Mac Pro 2013 lifecycle ( 2013-2022) , Apple sold more macs then ever. Made more money in the Mac product space than ever. What 'senses' are they suppose to come to. Make less money and become less successful?
IF Dell, HP , or Lenovo had much better margins then perhaps Apple would have to come to their senses and do a major change to tactics.
Apple not putting effort into selling systems into a relatively narrow subset of the market is not a matter of "senses" but of focus. Also two different viewpoints. ( "If Apple doesn't sell to me then it is an abject failure" . A somewhat narcissistic viewpoint. )
If affordable access to off-the-shelf GPU cards is your number one top priority (if primarily just looking for a container for a large , high power consuming GPU card), then yeah should switch to PC. The maximal power consuming GPU card is not likely going to be Apple's top priority going forward.
If Apple's next Mac Pro can handle a "raid" card with four x4 SSD MVNe PCI-e v4 ( x16 worth of PCI-e v4 bandwidth) and 3-4 x8 PCI-e v3/v2 cards they'd probably be in pretty good shape. Those are kinds of cards that TBv4/5 aren't really going to handle well in those aggregate configurations. They do not critically need PCI-e v5 or v6 to be a contender on this specific Mac Pro iteration. ( Similar 'doom' if not PCI-e v4 was trotted out months before MP 2019 introduction. Hasn't really made a huge dent in effectiveness over its first two years. )
