MP 7,1 Mac Pro 8.1 - WHAT IF...

[mode11]

Hmmm, I'll be honest with you, I'd LOVE for the 8.1 to be a vanity project. Just Apple going balls to the wall showing off Everything that they're capable of...cuz I'll tell ya what, that's a machine I wouldn't be able to wait to buy!

Yeah, but we want the Mac Pro to see continuous iterations. Not for them to throw everything at it once, then lose interest for the next six years. There needs to be a rational business case.

Personally, I'd like Apple to make the Mac equivalent of a desktop PC. That will never happen, so the next best thing is for a steady stream of workstation releases, so I can buy one second hand after a few years. If Apple had kept iterating on the original Mac Pro towers, I'd be happily using a ~2018 model right now. The massive gaps between releases, and the radical changes in form factor, are a PITA to be honest.

 

[singhs.apps]

Yeah, but we want the Mac Pro to see continuous iterations. Not for them to throw everything at it once, then lose interest for the next six years. There needs to be a rational business case.

Personally, I'd like Apple to make the Mac equivalent of a desktop PC. That will never happen, so the next best thing is for a steady stream of workstation releases, so I can buy one second hand after a few years. If Apple had kept iterating on the original Mac Pro towers, I'd be happily using a ~2018 model right now. The massive gaps between releases, and the radical changes in form factor, are a PITA to be honest.

This aspect of fiddling, chopping and changing things viz the Mac Pro doesn’t get enough attention.
It’s all about signalling and Apple has been found wanting when it comes to the Mac Pro.

This confused approach is, amongst other things, responsible for the weird state the Mac Pro finds itself in.

 

[mode11]

Making a workstation in itself is pretty straightforward. Apple have done it, and HP, Lenovo and Dell have no trouble either. The issue with Apple is that they'd rather not make a workstation. They've tried multiple times to kill it off, first with the 6,1, then putting a Xeon in a space-grey iMac. Even the 5,1 was only a minor revision of the 4,1. The 7,1 is a high-quality machine, but it seems the business case relies on charging a lot for it, then recouping development costs by selling it without changes for as long as they possibly can.

Perhaps this sort of machine is just not a good fit for Apple. Many of the corporate / education buyers who need fleets of workstations may want to run Windows software anyway (e.g. SolidWorks), and/or would appreciate CUDA-compatible video cards. Cost is also a factor when filling labs with these things - it adds up. PC OEMs actually have to compete with each other for large orders, so likely offer better discounts. This only leaves a relatively small number of creative professionals for whom macOS is a must.

 

[Mac3Duser]

I have a Windows workstation and never one BSOD. Windows has become much more stable than it was before.

 

[mode11]

Windows in a managed, network setting can be less fun. IT usually find a way of messing things up, whereas a personal machine can be well optimised.

I teach at a university, and a few years ago we got rid of our iMacs, replacing them with Alienware PCs (the uni had a contract with Dell) with GTX1080's and dual monitors. We do animation / VFX work, and Apple just didn't sell anything comparable for the money. The refusal to work with Nvidia is also pathetic, denying Mac users the option for real-time raytracing in apps like Blender and Unreal, if using an RTX card. Although I'm sure it's true most PC towers don't get upgraded in practice, our IT guys did add an extra 1TB SSD to each machine at one point, when the drives were filling up. Try doing that to an iMac - you'd need to unglue the screen first, even if there was a spare bay.

The Alienware's had USB issues, but that's another story.

 

[fuchsdh]

Making a workstation in itself is pretty straightforward. Apple have done it, and HP, Lenovo and Dell have no trouble either. The issue with Apple is that they'd rather not make a workstation. They've tried multiple times to kill it off, first with the 6,1, then putting a Xeon in a space-grey iMac. Even the 5,1 was only a minor revision of the 4,1. The 7,1 is a high-quality machine, but it seems the business case relies on charging a lot for it, then recouping development costs by selling it without changes for as long as they possibly can.

Perhaps this sort of machine is just not a good fit for Apple. Many of the corporate / education buyers who need fleets of workstations may want to run Windows software anyway (e.g. SolidWorks), and/or would appreciate CUDA-compatible video cards. Cost is also a factor when filling labs with these things - it adds up. PC OEMs actually have to compete with each other for large orders, so likely offer better discounts. This only leaves a relatively small number of creative professionals for whom macOS is a must.

I don't really think this is true, it's just that the Intel 7,1 was always intended to be a one-off while they got the AS Mac Pro out. They got the Intel transition done in well south of a year and while they gave themselves a looser timeline I'd bet somewhere in Apple there's a sad timeline where if not for all the supply chain and COVID issues the new Mac Pro is already out.

Now, delays aside, they should have just slapped the W-33XX Intel processors in the end of last year to keep people happy and make the line feel like it has a future (yes, that would have required another damn Intel socket change, but compared to the retooling they've got to do to go from that to the AS model it seems like a minor use of resources.) But I can understand that they originally expected they'd have the Intel lineup totally replaced by this point. The fact they explicitly mentioned the Mac Pro as the machine remaining back at WWDC I think indicates how much things slipped.

 

[Joe The Dragon]

I don't really think this is true, it's just that the Intel 7,1 was always intended to be a one-off while they got the AS Mac Pro out. They got the Intel transition done in well south of a year and while they gave themselves a looser timeline I'd bet somewhere in Apple there's a sad timeline where if not for all the supply chain and COVID issues the new Mac Pro is already out.

and the test systems with one pci-e slot seem like some plan failed or that the gen 1 chips did not have what was needed to replace the 7.1.
In ram / cpu / gpu and io.

And the gen 2 chips due to supply issues made it so that no mac pro in 2022.

But how long to wait?? and why not have an preview of what it will have with an order now?

Is the mac pro base pricing plan way to high and they don't want to be stuck with lot's of systems with an base ram of 256GB-512GB that start at $6000-$7000?
Looking for ways to have systems that may have ways to flex if you need more cpu then gpu + ones that where you may need more GPU then cpu?

Trying to find an way to have 4-8+ TB buses that all need to have video out? even when most users may only need 1-3 TB buses with video and maybe 3-6 ones that are just for data only?

 

[fuchsdh]

and the test systems with one pci-e slot seem like some plan failed or that the gen 1 chips did not have what was needed to replace the 7.1.
In ram / cpu / gpu and io.

And the gen 2 chips due to supply issues made it so that no mac pro in 2022.

But how long to wait?? and why not have an preview of what it will have with an order now?

Is the mac pro base pricing plan way to high and they don't want to be stuck with lot's of systems with an base ram of 256GB-512GB that start at $6000-$7000?
Looking for ways to have systems that may have ways to flex if you need more cpu then gpu + ones that where you may need more GPU then cpu?

Trying to find an way to have 4-8+ TB buses that all need to have video out? even when most users may only need 1-3 TB buses with video and maybe 3-6 ones that are just for data only?

I certainly have no idea, and I doubt we'll have much of an indication of what the issues were until we see what form of computer they actually ship.

As for pricing, my expectation is I don't think it'll go above $5–6K as the entry pricing. Like with the 7,1 I'm presuming that it's going to distinguish itself from the rest of the lineup by flexibility, even if it's just in terms of slots and being headless rather than the upgradable GPU, RAM, SSDs, etc. The flip side is if it's not offering as much flexibility I don't think they can really charge as much as they could for the 7,1 (and there just won't be as many crazy BTO specifications to boost the price, either.) The fact that they're selling the Ultra-equipped Studio at $4K means I doubt we get a Mac Pro model that low, unless they have some weird Max version. And who knows? As people are speculating if the Mac Pro is designed to sacrifice some of the SoC performance advantages for flexibility, it could be somewhat cheaper and the price just cranks up if you are trying to get equivalent performance. But of course why bet on Apple going low when they can go high (cost)? Apple might not be willing to take lower margins on specialty or low-volume chips if they're only getting used in the Mac Pro.

 

[mode11]

I'd bet somewhere in Apple there's a sad timeline where if not for all the supply chain and COVID issues the new Mac Pro is already out.

That's quite possibly true. It would make sense for them to prioritise hot-selling models if manufacturing / fabrication capacity is constrained. Plus for the sake of chip yields, you always start with the simple stuff then work upwards.

Now, delays aside, they should have just slapped the W-33XX Intel processors in the end of last year to keep people happy and make the line feel like it has a future (yes, that would have required another damn Intel socket change, but compared to the retooling they've got to do to go from that to the AS model it seems like a minor use of resources.)

I think at this point, they're not trying to raise the bar on the machine they need to replace. A new generation of Xeons, and perhaps 7000-series AMD cards, wouldn't make the comparison graphs look as impressive when the AS 8,1 is eventually announced.

Plus, it would reset the clock on Intel macOS support. You can't release a very expensive new machine at this point, then cut it from new OS releases after a couple of years. In fact, the risk of this makes it difficult to sell such a machine in the first place.

But I can understand that they originally expected they'd have the Intel lineup totally replaced by this point. The fact they explicitly mentioned the Mac Pro as the machine remaining back at WWDC I think indicates how much things slipped.

It wasn't a very explicit mention though. Only that it's a story for another day.

But how long to wait?? and why not have an preview of what it will have with an order now?

Yeah, whilst I get Apple's long-standing marketing MO of secrecy-followed-by-big-splash in the consumer space, it's awkward for businesses trying to plan purchases. Doubly so with the Mac Pro, since Apple's commitment to this category is suspect. They could just as easily decide to cancel it, due to the 'unexpected popularity' of the Studio. In fact, if the Studio really was a strong seller, they would be incredibly tempted.

Given the circumstances, I can't see what they have to lose by releasing key details about the new Mac Pro - it's not like anyone is going to copy them. Perhaps they're reserving the option to quietly not release it.

The flip side is if it's not offering as much flexibility I don't think they can really charge as much as they could for the 7,1 (and there just won't be as many crazy BTO specifications to boost the price, either.) The fact that they're selling the Ultra-equipped Studio at $4K means I doubt we get a Mac Pro model that low

At this rate, people will just go and buy a PC workstation. $4K would get a sweeeet PC. They're really testing people's commitment to macOS.

 

[maikerukun]

I don't really think this is true, it's just that the Intel 7,1 was always intended to be a one-off while they got the AS Mac Pro out. They got the Intel transition done in well south of a year and while they gave themselves a looser timeline I'd bet somewhere in Apple there's a sad timeline where if not for all the supply chain and COVID issues the new Mac Pro is already out.

Now, delays aside, they should have just slapped the W-33XX Intel processors in the end of last year to keep people happy and make the line feel like it has a future (yes, that would have required another damn Intel socket change, but compared to the retooling they've got to do to go from that to the AS model it seems like a minor use of resources.) But I can understand that they originally expected they'd have the Intel lineup totally replaced by this point. The fact they explicitly mentioned the Mac Pro as the machine remaining back at WWDC I think indicates how much things slipped.

In my view, they're not going to release until it satisfies people like me; power hungry, wattage consuming, extreme artists who need GPUs strong enough that we don't care what Nvidia is up to. It can't just be really good CPU's and middling GPU. It is called the PRO and has to have GENUINE PRO PERFORMANCE otherwise as has been mentioned in this thread, they will actually start losing the most unique corner in their usersphere.

 

[maikerukun]

At this rate, people will just go and buy a PC workstation. $4K would get a sweeeet PC. They're really testing people's commitment to macOS.

And THIS is the part that they're screwing up right now. Never in my life would I consider getting a PC as a MAIN in my HOME STUDIO.

NEVER.

And yet, here I am, genuinely contemplating picking up that Puget System with the dual 4090's in it. This is what I imagine my great (to the 10th power) grandmother Eve must've felt in the garden tempted by that snake, the devil (here represented by Puget Systems).

I don't want to eat the apple, but Apple is actively trying to leave me no choice SMH.

 

[avro707]

I have a Windows workstation and never one BSOD. Windows has become much more stable than it was before.

Win 11 for Workstations is very stable in my Mac Pro 7,1, but windows on my work laptop (managed environment by IT) is just a terrible pain. Every day it had some new way to frustrate me.

Looks like the 7,1 is my last Mac Pro unless they can make something Apple Silicon work with windows and take 3rd party GPUs.

 

[mattspace]

Plus, it would reset the clock on Intel macOS support. You can't release a very expensive new machine at this point, then cut it from new OS releases after a couple of years. In fact, the risk of this makes it difficult to sell such a machine in the first place.

I would bet money that Intel machines will continue to have new macOS support for longer than the first generation Apple Silicon Macs. The Mac Mini and Mac Pro will still be "new" that-year machines as long as Apple are selling them.

 

[mode11]

How did that work out for the Mac Pro 6,1?

 

[mattspace]

How did that work out for the Mac Pro 6,1?

3 years from when it was last on sale new, and 9 years after it was introduced and never progressed from 9 year old hardware. The OG M1 will go obsolete super-quickly, because it's the most marginal hardware specs - effectively it's an iPad in a different case.

 

[JayKay514]

Making a workstation in itself is pretty straightforward. Apple have done it, and HP, Lenovo and Dell have no trouble either. The issue with Apple is that they'd rather not make a workstation. They've tried multiple times to kill it off, first with the 6,1, then putting a Xeon in a space-grey iMac. Even the 5,1 was only a minor revision of the 4,1. The 7,1 is a high-quality machine, but it seems the business case relies on charging a lot for it, then recouping development costs by selling it without changes for as long as they possibly can.

Perhaps this sort of machine is just not a good fit for Apple. Many of the corporate / education buyers who need fleets of workstations may want to run Windows software anyway (e.g. SolidWorks), and/or would appreciate CUDA-compatible video cards. Cost is also a factor when filling labs with these things - it adds up. PC OEMs actually have to compete with each other for large orders, so likely offer better discounts. This only leaves a relatively small number of creative professionals for whom macOS is a must.

I don't know that they'd "rather not" do it, I think they'd rather not make a commodity Intel / AMD box that anyone with enough money and time can build for less.

The Mac Pro is a halo product, for better or worse, and for them to kill it off is to cede ground in the competitive space. But you do run into the issue of the tradeoff of macOS simplicity vs hardware lock-in.

But, looking at the 7.1 specifically, it's a big compromise, and if I were Apple I'd also want to obsolete it ASAP.

• Xeons with PCIe 4.0 / 5.0 were not shipping at the time they were in procurement for the launch, which meant a speed and width bottleneck on the expansion bus.
• Apple came up with MPX as a workaround. MPX is PCIe + nonstandard extra connection lanes using an internal Thunderbolt bus, with an extension socket added to the standard PCIe x16 slot. It is neat in that it can pass video internally to the main TB card output without the kludge of using in-out DP cables like many PC motherboards, IIRC.
• That said, MPX is not an industry standard. Nobody except Apple, AMD and maybe a handful of companies make things that work with it. It only works in this one specific computer. It's a microscopic niche market.
• In addition, Apple doesn't want to be stuck behind Intel's timetables. Not only is AMD lapping Intel in terms of tick-tock die shrinks and lower energy usage with better specs (notably, *way* more PCIe lanes on Epyc and Threadripper Pro), Apple and TSMC have already hit even smaller die feature sizes and lead in power-per-watt. Why would you trade that advantage away?
• So with the 7.1, Apple got stuck in a technological dead end - they can't fix things with a firmware upgrade, so Apple Silicon is the path out.

That said, with the SOC approach Apple has taken with Apple Silicon, the entire CPU and graphics system is on a unified die, and this can simplify the motherboard design considerably, and open up new possibilities.

• In theory, a future AS Mac Pro motherboard would mostly just interface the SOC with the I/O, onboard hardware and any PCIe slots, lowering cost.
• That could open the path to a socketed AS SOC that can be upgraded for pro users and a motherboard controller (based on a T3 or something) that can be flash upgraded if needed. I wouldn't hold my breath for it, but it's a possibility.
• The modularity of Apple Silicon re-opens some very interesting product paths, like expandable AS blade servers for rendering and compute-intensive scientific tasks. Blades could have specialized processors with more GPU cores, efficiency cores or power cores depending on the task, and you could use a mix of them seamlessly.
• This gives Apple a way to make inroads into the data center space, where ARM processors are seeing success with offerings from Ampere, HP, and others. Reducing energy usage in these environments is critical, and Apple could lead here.

There's still two hurdles to overcome:

• The actual physical RAM limit is still relatively low. The 7.1 can have 1.5 terabytes of RAM, the Mac Studio maxes out at 128GB. While Apple Silicon uses cool swapping tricks to get around this, for people that crunch huge data sets, or want to avoid any lag in streaming data from disk, they'll have to find a way to increase this.
• User expandability. Yes, at least a few x16 PCIe 5.0 slots would be awesome. But equally as critical are SSDs. If Apple needs magic high-performance nonstandard SSDs for critical system tasks, so be it, but it'd be nice to match high-end PC motherboards which have multiple M.2 PCIe slots for secondary storage / backups etc, so we don't have to have an endless nest of cables for external drives.

 

[Joe The Dragon]

• This gives Apple a way to make inroads into the data center space, where ARM processors are seeing success with offerings from Ampere, HP, and others. Reducing energy usage in these environments is critical, and Apple could lead here.

There's still two hurdles to overcome:

• The actual physical RAM limit is still relatively low. The 7.1 can have 1.5 terabytes of RAM, the Mac Studio maxes out at 128GB. While Apple Silicon uses cool swapping tricks to get around this, for people that crunch huge data sets, or want to avoid any lag in streaming data from disk, they'll have to find a way to increase this.
• User expandability. Yes, at least a few x16 PCIe 5.0 slots would be awesome. But equally as critical are SSDs. If Apple needs magic high-performance nonstandard SSDs for critical system tasks, so be it, but it'd be nice to match high-end PC motherboards which have multiple M.2 PCIe slots for secondary storage / backups etc, so we don't have to have an endless nest of cables for external drives.

As for server stuff the storage needs to change to

1. Raid 1 / Raid 5 / Raid 10 / Raid 6 hot swapping of blades + maybe just JBOD choices.
2. FULL DFU mode access from the IPMI
3. NO RAID 0 only on the boot / main os disks
4. Open Compute Project NIC slot / pci-e slots
5. pcie storage (non apple)
6. more ram

 

[maikerukun]

I would bet money that Intel machines will continue to have new macOS support for longer than the first generation Apple Silicon Macs. The Mac Mini and Mac Pro will still be "new" that-year machines as long as Apple are selling them.

I'll tell you what, I sure do hope you're right. Cuz I may not be getting off of this 7.1 bus anytime soon if that new Mac PRO doesn't literally blow me out of the water...

 

[fuchsdh]

I would bet money that Intel machines will continue to have new macOS support for longer than the first generation Apple Silicon Macs. The Mac Mini and Mac Pro will still be "new" that-year machines as long as Apple are selling them.

I don't expect there to be an appreciable difference. The M2 isn't that much more advanced than the M1, and they're still selling M1s and will be into 2023, if not potentially later (Cook loves his zombie Macs.) The M1s are also much better machines than the original intel Macs were (which is saying something, considering how much better they were than the outgoing G4 and G5s they replaced.) And they're still selling the Mac mini and Mac Pro. So I don't think it'll be the same story of the PPC->Intel transition.

The 2013 Mac Pro I expect they're going to drop as soon as they can, because hey it's a 2013 product. But I expect there's not going to be any more accelerated a timetable for Macs than there has been based on their architecture.

(Plus, if past performance is an indication, if you really want you can just run an unsupported OS on your Mac and there will be basically no problem. Even the performance hit is surprisingly marginal depending on the machine you're using, and while I expect the Intels will age worse, it's always a tradeoff of support, stability, and features for every consumer.)

 

[prefuse07]

I don't expect there to be an appreciable difference. The M2 isn't that much more advanced than the M1, and they're still selling M1s and will be into 2023, if not potentially later (Cook loves his zombie Macs.) The M1s are also much better machines than the original intel Macs were (which is saying something, considering how much better they were than the outgoing G4 and G5s they replaced.) And they're still selling the Mac mini and Mac Pro. So I don't think it'll be the same story of the PPC->Intel transition.

The 2013 Mac Pro I expect they're going to drop as soon as they can, because hey it's a 2013 product. But I expect there's not going to be any more accelerated a timetable for Macs than there has been based on their architecture.

(Plus, if past performance is an indication, if you really want you can just run an unsupported OS on your Mac and there will be basically no problem. Even the performance hit is surprisingly marginal depending on the machine you're using, and while I expect the Intels will age worse, it's always a tradeoff of support, stability, and features for every consumer.)

Yeah, the 6,1 has already been dropped per Ventura:

 

[mode11]

The Mac Pro is a halo product

It's also a specialised one. It's not necessarily an aspirational product that sells stuff lower in the range. Someone buying a Dell XPS laptop, for instance, isn't secretly lusting after (or possibly even aware of) the Precision 7865 tower.

Xeons with PCIe 4.0 / 5.0 were not shipping at the time they were in procurement for the launch, which meant a speed and width bottleneck on the expansion bus.

An RTX 4090 is barely affected by running on PCIe 3.0 vs 4.0, and doesn't even support 5.0. So calling PCIe 3.0 a bottleneck is overstating it a bit. Sure, you can get the same bandwidth with half the lanes on 4.0, but the MP 2019 has plenty of lanes.

MPX is PCIe + nonstandard extra connection lanes using an internal Thunderbolt bus, with an extension socket added to the standard PCIe x16 slot. It is neat in that it can pass video internally to the main TB card output without the kludge of using in-out DP cables like many PC motherboards, IIRC.

How useful is TB routing in practice though? Don't most people just plug their monitor(s) into their graphics card?

In theory, a future AS Mac Pro motherboard would mostly just interface the SOC with the I/O, onboard hardware and any PCIe slots, lowering cost.

Using what? Where's the PCIe lanes on M1/2?

The modularity of Apple Silicon re-opens some very interesting product paths, like expandable AS blade servers for rendering and compute-intensive scientific tasks.

That's an interesting idea.

This gives Apple a way to make inroads into the data center space, where ARM processors are seeing success with offerings from Ampere, HP, and others.

Apple tried this sort of thing with the Xserve. They like to add value with design, and stuff that goes in a server room doesn't give them much opportunity to shine in this regard. More to the point, would Apple be regularly iterating on server chips, with a published long-term roadmap? Or would they just lose interest after a few years? Would these chips have much relevance to their consumer products? That sort of ARM stuff is better left to Ampere.

 

[innerproduct]

Semi-relevant: first geekbench numbers for 7900xtx cards. https://browser.geekbench.com/v5/compute/6015593
Basically 50% faster than the 6900XT for about the same power. Early numbers and for some tasks it might be faster or slower. At least it is an indication.
So by now we know the playing field for high end workstations for the coming two years. Intel Sapphire falls based on golden cove up to 56 cores and massive amounts of PCI and RAM. Then we have current thread ripper pro and upcoming 7* series that is a bit faster. Also with massive amounts of PCI lanes and ram.
A mid range VFX workstation from Dell or HP will use these, coupled with at most 2 pro RTX 6000 or optionally some prosumer card like the above 7900xtx or nvidia 4090.

I was thinking....
If I was responsible at Apple, I would have identified markets where Macs have weaknesses and assembled a pro team to remedy this. Really understand the needs and workflows pros have. Maybe employ a lot of people to help companies to get up to speed a deliver great solutions on the hardware that is upcoming....
To be honest, isn't that what they have done since 2017 with the pro workflow team? Maybe it takes a little longer than anticipated, but all in all, apple have step by step delivered on everything they said even if they are not superfood at being transparent in their communication . Video editing is more or less a solved issue on the Mac these days. All major 3d companies seems to have gotten help from apple to optimise their software. From Octane and Redshift through Autodesk and Sidefx etc.
Maybe, just maybe, apple is on top of this, just a few months late and in March next year we will have systems that are really outstanding. Actually, to me, that sounds like the most plausible outcome. Maybe not with HW that beats the last single percent of custom cooled multi GPU render machines, but something better than most of the competition.

feeling hopeful today

 

[mode11]

For those that lived through the PowerPC era, this is mildly concerning. That was also a power-efficient architecture that (essentially) only Apple used, which came out of the gate strong, outperforming x86 clock-for-clock with it's smarter way of doing things. Then x86 got its act together and pulled way ahead.

Obviously things are different now, as the success of the iPhone means Apple can invest a lot of money each year in developing its industry-leading SoC, which certainly scales up nicely to laptop level (including desktops like the mini, Studio and iMac that use a laptop chip or two). It's unproven in the workstation space as yet though.

It needs a version that allows the connection of multiple GPU units - perhaps like the blade system @JayKay514 described. Although that does sound suspiciously like a desktop PC, with multiple graphics cards plugged into a PCIe bus - only doubtless much more expensive. Plus, are Apple really better at designing GPUs than NVIDIA? Or even AMD? It's not really a core competency, if you'll excuse the pun.

 

[JayKay514]

It's also a specialised one. It's not necessarily an aspirational product that sells stuff lower in the range. Someone buying a Dell XPS laptop, for instance, isn't secretly lusting after (or possibly even aware of) the Precision 7865 tower.

True. But it does give them bragging rights which extends to the other M-series chips.

An RTX 4090 is barely affected by running on PCIe 3.0 vs 4.0, and doesn't even support 5.0. So calling PCIe 3.0 a bottleneck is overstating it a bit. Sure, you can get the same bandwidth with half the lanes on 4.0, but the MP 2019 has plenty of lanes.

Well, Apple seemed to feel it was necessary to build those dual-GPU MPX modules. If the Xeon had supported PCIe 4.0 it might not have been necessary. I'm just theorizing, of course!

How useful is TB routing in practice though? Don't most people just plug their monitor(s) into their graphics card?

Usually, yes, but if you're daisy-chaining stuff via TB4 it helps reduce cable clutter.

Using what? Where's the PCIe lanes on M1/2?

Aye, there's the rub. That's why either the SOC will need to have more of an onboard PCIe infrastructure, or hand that off to the motherboard chipset and have some sort of wide-bandwidth connection.

Apple tried this sort of thing with the Xserve. They like to add value with design, and stuff that goes in a server room doesn't give them much opportunity to shine in this regard. More to the point, would Apple be regularly iterating on server chips, with a published long-term roadmap? Or would they just lose interest after a few years? Would these chips have much relevance to their consumer products? That sort of ARM stuff is better left to Ampere.

You might be right, but I think Apple Silicon positions them a bit differently now. The old G3/G4 Xserve wasn't really of interest to anyone except existing MacOS users, and the current rackmount Mac Pro is definitely not cost-efficient for large hosting / data center operations.

General-purpose ARM chips may excel at average server tasks, but what the M-series has going for it is a lot of dedicated onboard GPU units, AV encode / decode hardware, and very low power requirements. Probably overkill for web servers, but very much of interest for any company that handles media uploads / capture, 3D rendering, cloud streaming, etc.

Also consider, if/when Apple comes out with its AR / VR consumer solution, they will want to support content creators and app developers with complementary hardware. Maybe this won't be sold, per se, but will run in Apple datacenters as a hosted solution. For instance, being able to do game streaming for the Apple TV platform, including smart TVs that run the Apple TV app.

 

[ZombiePhysicist]

Usually, yes, but if you're daisy-chaining stuff via TB4 it helps reduce cable

Thunderbolt causes a hell spawn of wiring. It doesn’t reduce cable clutter. It is quite literally the spawn of cable clutter. Many a trashcan Mac have illustrated this.