MP All Models Mac Pro X,Y…WHAT IF…

[mattspace]

What might have been.
Imagine MacOS on one of these.
No need to reinvent to reinvent the wheel for best in class performance.

Dell Precision 7875 Review: 96-Core Threadripper Pro and Dual RTX GPU Powerhouse

The Dell Precision 7875 is a highly impressive high-end workstation with an AMD Ryzen Threadripper Pro 7000WX series CPU.

www.storagereview.com

The single GPU HP Z6 seems to match, or outpunch that Dell in most things, according to the testing shown.

 

[Harry Haller]

The single GPU HP Z6 seems to match, or outpunch that Dell in most things, according to the testing shown.

That's the great thing about vendor choice.
You can go with whatever suits your needs.
Thankfully there are smaller ones like Boxx and Puget for even more price ranges than the Dells and HPs of the world.

Nevertheless, I'm still holding out the faintest of hope for a truly modular and upgradable M3 or M4 Extreme in the next year or two.

 

[avkills]

In the mean time, E1.S drives are becoming an appliance thing, but Apple is nowhere:

https://twitter.com/i/web/status/1759760199429001510

I might be getting one of those! Thank you for the link. That is if work agrees to buy.

 

[deconstruct60]

In the mean time, E1.S drives are becoming an appliance thing, but Apple is nowhere:

https://twitter.com/i/web/status/1759760199429001510

The following Seems to be lacking M-series drivers, but that is 'fixable' by the vendors ( Apple isn't particularly blocking this (hence working in macOS 13.x on Intel side .. just need to get past trying to do it with kernel extensions; that is a dead end. )

https://www.highpoint-tech.com/product-page/ssd7749e-311r1c

6-pin power in MP 2023. Check. Full length, dual width. Not a problem.
Drop the hardware RAID aspect and it would likely work 'out of the box' with standard Apple supplied NVMe drivers.
The 'hold up' here is likely trying to present as a non standard conforming drive.

For example 8x M.2 NVMe card works just fine in MP 2023

M.2 8x4 Silent Gen4 PCIe Card - Tech Specs - SONNETTECH

View videos, reviews and tech specs for Sonnet M.2 8x4 Silent Gen4 PCIe Card and determine hardware and OS compatibility.

www.sonnettech.com

U.2 drive no "hardware RAID layer" ... works.

Fusion Dual U.2 SSD PCIe Card - Tech Specs - SONNETTECH

View videos and tech specs for Fusion Dual U.2 SSD PCIe Card, and determine hardware and OS compatibility.

www.sonnettech.com

 

[ZombiePhysicist]

The following Seems to be lacking M-series drivers, but that is 'fixable' by the vendors ( Apple isn't particularly blocking this (hence working in macOS 13.x on Intel side .. just need to get past trying to do it with kernel extensions; that is a dead end. )

https://www.highpoint-tech.com/product-page/ssd7749e-311r1c

6-pin power in MP 2023. Check. Full length, dual width. Not a problem.
Drop the hardware RAID aspect and it would likely work 'out of the box' with standard Apple supplied NVMe drivers.
The 'hold up' here is likely trying to present as a non standard conforming drive.

For example 8x M.2 NVMe card works just fine in MP 2023

M.2 8x4 Silent Gen4 PCIe Card - Tech Specs - SONNETTECH

View videos, reviews and tech specs for Sonnet M.2 8x4 Silent Gen4 PCIe Card and determine hardware and OS compatibility.

www.sonnettech.com

U.2 drive no "hardware RAID layer" ... works.

Fusion Dual U.2 SSD PCIe Card - Tech Specs - SONNETTECH

View videos and tech specs for Fusion Dual U.2 SSD PCIe Card, and determine hardware and OS compatibility.

www.sonnettech.com

There is an entire thread on this here, but the bottom line is nothing modern works.

The question is which of these work that are bootable. The sonnet works, but basically only with the Micron 9300 Pro drive. Other more modern drives do not work in the sonnet.

The problem persists for the HighPoint cards. They work, but only with older U.2 drives and a very limited number of them. Or if you marry multiple M.2 sticks, it works, but isn't bootable.

For example, the Micron 9400 Pro 30TB drive does not work with either of those cards.

There are other cards that let you piece together over 15TB using M.2 sticks, however, their problem is, while they will work and show up on macOS, they are not bootable (at least not pooled so the largest bootable drive you could have there is a JBOD single stick of 8TB).

Ergo, the problem is that something in macOS is incompatible with new U.2/U.3/EDSFF type drives, and seemingly purposefully so, and the last and largest single bootable SSD you can get is the 15tb Micron 9300 Pro (outside ancient spinning SATA drives). No other larger bootable SSDs exist for macOS. All these drives work with zero problems on the very same hardware booting into windows.

Apple borked support somehow. There are reports somewhere in that thread that some earlier versions of macOS specifically used to support more U.2 drives that are now incompatible.

It's a patchwork of randomness. For whatever reason, the Micron 9300 Pro continues to work, while other U.2 drives do not.

The current search on that thread is...are there any modern bootable U.2, U.3, EDSFF drives supported under macOS, and for months and months, if not years at this point, the answer has been a disheartening: nope.

 

[mattspace]

There is an entire thread on this here, but the bottom line is nothing modern works.

The thing about not being able to boot from RAID wouldn't be so bad, if you could easily map the default user directories - documents, music, movies etc to an external storage system (so you can keep the custom icon for finder's sidebar etc).

It is beyond stupid that Apple simultaneously don't support larger onboard storage (than you can get in a laptop), don't make a NAS/DAS, and don't provide a turnkey UI for remapping those privileged directories.

Build a first party storage array, or make USB / TB / SMB volumes work indistinguishably from onboard storage *mutter*.

 

[ZombiePhysicist]

The thing about not being able to boot from RAID wouldn't be so bad, if you could easily map the default user directories - documents, music, movies etc to an external storage system (so you can keep the custom icon for finder's sidebar etc).

It is beyond stupid that Apple simultaneously don't support larger onboard storage (than you can get in a laptop), don't make a NAS/DAS, and don't provide a turnkey UI for remapping those privileged directories.

Build a first party storage array, or make USB / TB / SMB volumes work indistinguishably from onboard storage *mutter*.

I’ve tried mapping in the past but it always leads to sadness. At some point one of the drives won’t mount, the dangling link either leads to new now unsynchronized libraries and/or data loss. If anything, things have gotten worse on Apple hardware with unmounting. So never again will that be an option for me. As always, ymmv.

 

[mattspace]

I’ve tried mapping in the past but it always leads to sadness. At some point one of the drives won’t mount, the dangling link either leads to new now unsynchronized libraries and/or data loss. If anything, things have gotten worse on Apple hardware with unmounting. So never again will that be an option for me. As always, ymmv.

yeah, I still have all my 4,1's storage in a JBOD attached to a 2012 mini, and shared via SMB, while I try to figure out my onboard storage in the 7,1. Despite having the drives set to mount on login, every reboot I have to manually add them all to Locations in the finder sidebar (and obviously, I have to manually wake the mini server before waking the 7,1, because the 7,1 won't reliably wake the mini when seeking network-based resources).

 

[danwells]

That photographer’s workflow was pretty extreme.

Lloyd Chambers is pretty well known in the photographic community for his grousing. He spots legitimate problems, but he ALSO deliberately goes for edge cases that nobody else ever encounters. I haven't heard ANY other reports of crazy fan noise from a 16" MBP. Some of his homemade tools are egregious power-virus tests similar in concept to what FurMark was some years ago. Sure, you can cook a powerful MacBook Pro with some of his tools, but you can build an egregious tool that will cook a Cray...

 

[mode11]

I'll just leave this here...

They have to spec the PC up with extravagant upgrades like 16TB of SSDs and 2x RTX4090's to get the PC to the price of a Mac Pro. Needless to say, it tears the Mac a new a*shole.

 

[mattspace]

I'll just leave this here...

They have to spec the PC up with extravagant upgrades like 16TB of SSDs and 2x RTX4090's to get the PC to the price of a Mac Pro. Needless to say, it tears the Mac a new a*shole.

You know the thing that really kills me is for half the price of a Mac Pro, I can buy a midrange Lenovo P Series - and get the full industrial grade workstation experience - huge ECC RAM capacity, pro graphics, Xeon processors, like it doesn't mater if it's not even as fast at purchase as a Mac Pro, I rarely if ever use the top performance, I just want the performance I have to be from workstation-class pro-reliable components. Not a scaled-up cellphone.

 

[mode11]

I just want the performance I have to be from workstation-class pro-reliable components. Not a scaled-up cellphone.

You may be on the wrong platform (and I say that with all sympathy). It doesn't have to be this way, but is the course Apple has chosen for now.

Whilst Apple would obviously prefer to have one architecture, for numerous practical and marketing reasons, there might have been a case for their workstations remaining on x86. The Max chips would have been fine for top end Studio / iMac models, and they could have avoided the need to make the Ultra (and Extreme, if that was ever on the cards). Whilst still having a (preferably Threadripper) workstation comparable anything on the PC side. As a point of comparison, it may be the case that Windows exists in both expandible-x86 and ARM-SoC forms for the long haul, with strengths to each.

Once the MP switched to AS in 2023, though, that possibility disappeared. It would look bizarre to switch the MP back to x86 in the foreseeable future. It would destroy whatever credibility Apple has left in the workstation market, and reflect badly on the potential for Apple Silicon too, sending the message that not only can it not cut it at the high end, it likely never will. Though admittedly, Apple already has close to zero credibility in the workstation market, and it's pretty obvious a doubled-up laptop SoC is never going to make a credible workstation, so perhaps the only loss would be face.

 

[Harry Haller]

You know the thing that really kills me is for half the price of a Mac Pro, I can buy a midrange Lenovo P Series - and get the full industrial grade workstation experience - huge ECC RAM capacity, pro graphics, Xeon processors, like it doesn't mater if it's not even as fast at purchase as a Mac Pro, I rarely if ever use the top performance, I just want the performance I have to be from workstation-class pro-reliable components. Not a scaled-up cellphone.

Oh snap.

AS for watches…brilliant.
AS for phones…brilliant.
AS for tablets…brilliant.
AS for laptops…brilliant.
AS for Minis and Mac Studios…brilliant.
AS for HEDT and workstations…epic fail.

 

[danwells]

Apple's expandable workstation market is tiny - not worth maintaining a second build of MacOS for (and forcing all the creative pro app companies to optimize THEIR builds for two architectures). Apple sells something well north of 10 million Macs annually (I haven't found an annual figure, but they sometimes sell over five million in a quarter). How many would be an expandable workstation starting over $10,000 (it would have to to cover the expense of an extra build, plus the appealing CPUs are in the $3000-$10,000 range before building a computer around them)? 50,000? 100,000 would be a lot.

The second problem is that an M3 Ultra is going to embarrass a lot of very big workstation processors for a lot of tasks. An M3 Extreme, were such a thing to exist, will embarrass almost any workstation processor for most tasks. The ONLY markets that would prefer a Threadripper are those that need a ton of RAM and those that need multiple GPUs. An M3 Ultra will be pretty competitive with a desktop 4090 or one of the higher-end "standard" workstation GPUs (not the crazy compute accelerators that cost tens of thousands of dollars). An M3 Extreme should beat anything EXCEPT a very high-end compute accelerator or a bunch of GPUs.

It's probably easier for developers to build in support for back-end compute farms as necessary than to keep the Intel (or AMD) architecture alive for that small market. If you need massive compute acceleration or terabytes of RAM, build a headless box that does that and connect your Mac to it over multiple Thunderbolt and/or 10 Gb Ethernet links.

Crays rarely supported direct consoles except for maintenance. You connected to them with another computer that ran the UI, sometimes using super-bandwidth links...

 

[ZombiePhysicist]

Apple's expandable workstation market is tiny - not worth maintaining a second build of MacOS for (and forcing all the creative pro app companies to optimize THEIR builds for two architectures). Apple sells something well north of 10 million Macs annually (I haven't found an annual figure, but they sometimes sell over five million in a quarter). How many would be an expandable workstation starting over $10,000 (it would have to to cover the expense of an extra build, plus the appealing CPUs are in the $3000-$10,000 range before building a computer around them)? 50,000? 100,000 would be a lot.

The second problem is that an M3 Ultra is going to embarrass a lot of very big workstation processors for a lot of tasks. An M3 Extreme, were such a thing to exist, will embarrass almost any workstation processor for most tasks. The ONLY markets that would prefer a Threadripper are those that need a ton of RAM and those that need multiple GPUs. An M3 Ultra will be pretty competitive with a desktop 4090 or one of the higher-end "standard" workstation GPUs (not the crazy compute accelerators that cost tens of thousands of dollars). An M3 Extreme should beat anything EXCEPT a very high-end compute accelerator or a bunch of GPUs.

It's probably easier for developers to build in support for back-end compute farms as necessary than to keep the Intel (or AMD) architecture alive for that small market. If you need massive compute acceleration or terabytes of RAM, build a headless box that does that and connect your Mac to it over multiple Thunderbolt and/or 10 Gb Ethernet links.

Crays rarely supported direct consoles except for maintenance. You connected to them with another computer that ran the UI, sometimes using super-bandwidth links...

The Cray isn't the model. It ignores that a workstation class market very much does exist, on windows and linux, and has its purposes. There are classes of applications where you just need the horsepower local. I do agree, farming it out makes a great amount of success for many, but not for all. And forcing people into a segment of the market act doesnt make sense for them ignores those realities.

That said, your earlier statements on the Ultra/Extreme give me a little bit of hope for actually servicing at least some percentage of the workstation market that mac people still would like. I think you are right, not about the ultra, but the extreme. The ultra, may compete with older cards, but with the latest and greatest, I'm not sure it will get there. The Extreme might.

Don't get me wrong, I would be beyond pleased if the ultra could get there by itself. Unfortunately, my best guess is we are waiting till late fall for some Mac Pro announcement, if it were to come at all. I'd be very pleased to be wrong about that and hear earlier.

 

[mattspace]

The second problem is that an M3 Ultra is going to embarrass a lot of very big workstation processors for a lot of tasks.

If I recall, the M2 Ultra got itself embarrassed by Intel's consumer i9 processors in relatively short order. I expect the M3 Ultra to have exactly the same trajectory. Nothing stays "the fastest" for its entire service life, let alone for its entire depreciation life, or its entire current model life.

 

[ZombiePhysicist]

If I recall, the M2 Ultra got itself embarrassed by Intel's consumer i9 processors in relatively short order. I expect the M3 Ultra to have exactly the same trajectory. Nothing stays "the fastest" for its entire service life, let alone for its entire depreciation life, or its entire current model life.

It seems the M2 Ultra got embarrassed by the M3max itself, and in a very few months. Frankly, even studio buyers probably feel dumb. Why bother. Just get the laptop. The Extreme would make a much better case for a workstation. Or if an outright extreme isn't doable, why not just dual M3Ultras? I guess the ram and video might be a mess to coordinate in packages like that?

 

[mattspace]

It seems the M2 Ultra got embarrassed by the M3max itself, and in a very few months. Frankly, even studio buyers probably feel dumb. Why bother. Just get the laptop. The Extreme would make a much better case for a workstation. Or if an outright extreme isn't doable, why not just dual M3Ultras? I guess the ram and video might be a mess to coordinate in packages like that?

The point of the Ultra is display support (and 2 more TB ports / busses) - 6 displays, vs. usually 4 or fewer on the lower order processors. Again like the Mac Pro itself, performance isn't the point, it's the flexibility of configuration - how octopus-y it can be.

 

[ZombiePhysicist]

The point of the Ultra is display support (and 2 more TB ports / busses) - 6 displays, vs. usually 4 or fewer on the lower order processors. Again like the Mac Pro itself, performance isn't the point, it's the flexibility of configuration - how octopus-y it can be.

Well I think that's certainly some part of it. But I think a lot of people are hoping for a good boost in both CPU and GPU performance. To be fair it does that too. Just not enough. When it's own 'dominance' is eclipsed a few months alter by a much lower grade of processor, it feels like somewhat of a foolhardy exercise. I grant you, if the extra octopus part of the chip is your main focus, you do still have that.

 

[mattspace]

Well I think that's certainly some part of it. But I think a lot of people are hoping for a good boost in both CPU and GPU performance. To be fair it does that too. Just not enough. When it's own 'dominance' is eclipsed a few months alter by a much lower grade of processor, it feels like somewhat of a foolhardy exercise. I grant you, if the extra octopus part of the chip is your main focus, you do still have that.

yeah absolutely the performance of the Ultra is a thing - but whatever performance you pay a premium for on the top end will be eclipsed by the midrange in a year, what the midrange won't do is match the expansion capabilities.

But we've been dealing with that for ever - people have been saying "a macbook / iMac is faster in FCP than a Mac Pro" since 2014 or so.

 

[ZombiePhysicist]

yeah absolutely the performance of the Ultra is a thing - but whatever performance you pay a premium for on the top end will be eclipsed by the midrange in a year, what the midrange won't do is match the expansion capabilities.

But we've been dealing with that for ever - people have been saying "a macbook / iMac is faster in FCP than a Mac Pro" since 2014 or so.

Kind of. The Mac pros get neglected for a long time and I grant that’s the case for the trashcan. But the 2010 out punched over the iMacs for quite a while and so did the 2019. The 2019 wasnt eclipsed mere months later by a MacBook.

Particularly with expansion GPUs they stayed relevant for years. Heck, a dual 6800 duo setup still beats out m3 max and m3 ultra graphics performance years later.

And with the decided lack of that expansion, the studio getting soundly eclipsed a mere few months later, it makes it somewhat more of a greater sting. And to your point, all that’s left is the octopus factor. It’s somehow feels a lot more sharply futile a product.

 

[mode11]

Yeah, given neither have internal expansion, one might question the point of a Mx Ultra Studio, when a few months later you could buy a Mx+1 Max 16" MBP with similar performance and price. And of course have the XDR screen etc. thrown into the bargain. Though admittedly, the Studio would be quieter, and better suited to cranking through long encodes / renders.

 

[mode11]

Apple's expandable workstation market is tiny - not worth maintaining a second build of MacOS for (and forcing all the creative pro app companies to optimize THEIR builds for two architectures).

Sure, that ship has obviously sailed. Apple's workstation market was always small, and their AS architecture scales all over the rest of the Mac range (very well).

The second problem is that an M3 Ultra is going to embarrass a lot of very big workstation processors for a lot of tasks.

For CPU, sure. For GPU, not so much. According to Blender GPU benchmarks, the M2 Ultra (76c) gets 3220, the M3 Max (40c) gets 3460, the RTX4070 gets 5400, and an RTX 4090 gets 11280. So assuming perfect scaling, the M3 Ultra will approach an RTX4070 Ti Super (7230). Very impressive given the likely power consumption, but not really competitive with PCs, whether in cost at the low end or performance at the high end.

An M3 Extreme, were such a thing to exist, will embarrass almost any workstation processor for most tasks.

It's doubtful to ever exist though, given (as you say) the tiny Mac workstation market.

The ONLY markets that would prefer a Threadripper are those that need a ton of RAM and those that need multiple GPUs.

Oh OK, *just* those markets.

An M3 Ultra will be pretty competitive with a desktop 4090 or one of the higher-end "standard" workstation GPUs (not the crazy compute accelerators that cost tens of thousands of dollars).

Unlikely, see above.

An M3 Extreme should beat anything EXCEPT a very high-end compute accelerator or a bunch of GPUs.

Ditto.

It's probably easier for developers to build in support for back-end compute farms as necessary than to keep the Intel (or AMD) architecture alive for that small market. If you need massive compute acceleration or terabytes of RAM, build a headless box that does that and connect your Mac to it over multiple Thunderbolt and/or 10 Gb Ethernet links.

Seems like an expensive workaround, simply because it doesn't suit Apple's business model to build a proper desktop workstation. That's not the user's problem; they can just buy a PC instead.

Crays rarely supported direct consoles except for maintenance. You connected to them with another computer that ran the UI, sometimes using super-bandwidth links...

OK.

 

[mattspace]

Yeah, given neither have internal expansion, one might question the point of a Mx Ultra Studio, when a few months later you could buy a Mx+1 Max 16" MBP with similar performance and price. And of course have the XDR screen etc. thrown into the bargain. Though admittedly, the Studio would be quieter, and better suited to cranking through long encodes / renders.

6 displays, Vs. 4. 6 Thunderbolt ports, Vs. 4. That's the standout difference of things you absolutely cannot do with a Max processor Vs. an Ultra - you can let the render run a bit longer, you can go make coffee like in the old days etc, but the Max places absolute limits on what the hardware can be extended to do. Realistically, that's the only thing to lose sleep over, when trying to decide which one to get.

 

[mode11]

6 displays, Vs. 4. 6 Thunderbolt ports, Vs. 4. That's the standout difference of things you absolutely cannot do with a Max processor Vs. an Ultra - you can let the render run a bit longer, you can go make coffee like in the old days etc, but the Max places absolute limits on what the hardware can be extended to do. Realistically, that's the only thing to lose sleep over, when trying to decide which one to get.

What use case do you have for that many displays? I'm sure they exist, I just can't personally imagine needing more than the 3 displays I've got on my desk (my head only turns so far!). Video editors would want a TV as a reference monitor, of course, but not sure how much real estate one person can make use of otherwise.