I suspect the 2013 Mac Pro was sort of an experiment by Apple to test out the market for high end desktops. The radical redesign reflects where they think desktops are heading or at least where they see the market for Apple desktops heading. I think each design choice makes sense if you look at it from that perspective.

- Limited internal expansion / high speed external bus : In a future world where every computer has USB-C and/or ThunderBolt ports external expansion is going to be the obvious choice for most people. Very few individuals/companies use desktops exclusively so interoperability between laptops & desktops is another factor here. This is also true of internal storage. Fast internal storage + NAS/SAN are clearly the right direction to go here.

- Less CPU / more GPU : You can scale many highly parallel / CPU intensive tasks in the cloud or on a rack of your own servers better than you can buying the latest and greatest Mac Pro every few years. GPU processing is harder to scale outside of the box sitting on your desk. This doesn't meet everyone's needs today but we're talking about the future here. In 5-10 years are people going to be replacing their desktops often for more processing power? Some but probably not enough to make a high end desktop Mac a sustainable product for Apple.

- High base price : This is the real test. Anyone can say they want a high end Mac desktop but how many people are actually going to buy a $3k+ desktop? How often will they replace it? The price probably reflects the minimum amount of profitability Apple wants to continue making Mac Pros. Now it has sat on the shelf for 3 years. Are people still buying it? Like it or not that's a factor Apple has to consider. Can they sell old stock or do they have to update this thing every 6 months for it to be a viable product?

So if this was indeed an experiment we can only speculate about the results. Either the Mac Pro (in it's current form) is not a viable product for Apple or they're going back to the drawing board based on their experience with the 2013 model.

> how many people are actually going to buy a $3k+ desktop?

A high-end workstation is going to cost a lot more than that, particularly with Tesla or Xeon Phi coprocessors included. A single 7120A will set you back $4,841 and change.

The point being that the Mac Pro doesn't really satisfy, or target, the workstation crowd any longer.

I bought a Mac Pro in 2013 (the entry level model) and have been using it reliably for about 3 years.

With two 27 inch lcds, and a thunderbolt external drive, it's been solid development machine. I can also boot an external windows drive to play PC games, with ok performance on the last gen of games. The form factor is great for traveling (I can stuff it in a suitcase and take it anywhere).

Last few weeks it's been kernel panicking a couple times a day (when under heavy stress). And it'd be nice to have upgraded GPU to play the latest batch of PC games (and VR).

I'm pretty disappointed Apple has made zero updates to this model in over 1000 days. Given how much use I'd gotten out of this machine, it's hard to say it's been a mistake purchasing it. But there's also not really much I can do to upgrade it since external GPUs don't really work with Thunderbolt 2.

My guess is the Mac Pro line gets quietly discontinued by next year. 15 yrs ago I bought a G4 Cube, some lessons I never learn.

The existence of USB-C and Thunderbolt 3 means that the Mac Pro, as it is right now, might not be very necessary.

Processors and RAM are important, of course, if you're doing intensive work, but things like GPUs could be attached separately via Thunderbolt 3. I could see AMD creating a pro-level FirePro external TB3 "card", a box like an external hard drive that you just plug in via TB3 and off you go.

Apparently only one of the FirePro GPUs on the Mac Pro is hooked up in such a way as to contribute towards graphics performance, with the other being solely for compute; an external box designed specifically to hold a FirePro chip and associated hardware (VRAM, VMU, etc) could be styled, cooled, and shaped differently, and could provide a simple add-on for people who want parallel computation.

In other words, we could use TB3 to add on FirePro computation the way we use USB to add on hard drives.

With that kind of philosophy, the Mac Pro could be cheaper and easier to build, more expandable and not less, and appeal to more people. If I could order what was effectively a Mac Mini but with 32 GB of RAM and a Xeon or high-end desktop i7, I would buy ten of them tomorrow, and for people who want to do things like hardware-accelerated computation or transcoding they could buy add-ons that would do that work for them.

> Last few weeks it's been kernel panicking a couple times a day (when under heavy stress).

This used to happen to me as well(late August/early September). When it was under medium/heavy load it would freeze up occasionally and eventually kernel panic. I completely wiped the drive and reinstalled macOS which completely resolved the problem. I just chalked it up to a weird system file being corrupted.

> I can also boot an external windows drive

I didn't know this was possible. Does this work only on the Mac Pro?

Works on all Macs (tried on MacBook Air and Pro), the problem is convincing Windows to install on the external drive. If you have a Windows Pro license you could use Windows to Go and install it on an external drive.

It's not too hard with Windows 10 to do the Windows To Go process manually in UEFI mode. I've had pretty good luck with online guides.

The Mac Pro was well enough received by some, at the time. It really excels if you need beefy AMD GPUs (from a few years ago) and want something compact and quiet and good looking on your desk. The problem is that while it has pretty good performance in general, there are crucial bottlenecks for a number of different potential customers which make some other product a better fit.

For folks like software developers or anyone else who is bottlenecked by single-threaded CPU performance, the Mac Pro is no faster than a machine like an iMac, and not necessarily even much faster than a current laptop.

For folks like photo/graphics/video pros who care about the best large high-resolution display, previous-gen Thunderbolt connectors/protocols and external display hardware weren’t quite ready until the past year. Apple could hack around the problems with the iMac because everything was internal, but the problems hadn’t quite been solved for a desktop tower.

For folks needing beefy GPU performance, the market is segregated between those needing workstation class GPUs and ECC memory, etc., many of whom are running CUDA software that will only run on nvidia chips (oops), and those who want to play video games or similar, and would rather cut costs with consumer-grade hardware.

For folks worried about future-proofing a big purchase, there was a lot of uncertainty about ongoing product support and upgradeability.

And of course for anyone trying to cut costs, there are a variety of cheaper Windows PC towers optimized to particular use cases.

Etc.

Ultimately, for many customers (say, university computer labs, software developers, photographers, graphic designers, even pro audio and video people), an iMac makes a better desktop machine than the current form of the Mac Pro, whereas for the folks who need absolute max performance, it makes more sense to run their work on some server cluster or something. Pro creative software that used to run primarily on desktop workstations doesn’t necessarily need them anymore in the same way. Video game players who make up much of the PC tower market are happy to stick with Windows.

It’s not that easy to find a cohesive market of folks who really need a super-expandable mid-to-high-end Mac desktop tower, and sell a specific bundle of features to them.

If they release a new Mac Pro within the next year or so, with the same concept/chassis as the first one but upgraded internals, it might become a reasonable fit for some of those customers, or they might have been chased away by now, it’s hard to say.

This bit from the OP also rings true:

> In this case, Apple’s timing appears to have been bad. There’s no way to know for sure, but it seems like Apple decided to dance to its own rhythm, skip an Intel chip generation, and then wait for the good stuff, only to get bitten by slippage in Intel’s schedule. If that’s what happened, it’s hard to blame it on Intel. After all, it was Apple’s gamble.

I think if there had been a couple of additional Mac Pro hardware updates since the original release people would be substantially less pessimistic about it now.

To stay relevant I think they would have to support a dual-cpu configuration, and offer Tesla and Xeon Phi cards as options. High end RAM and disk setups would still be difficult to support in the current form factor though.

I'd love to see someone start building modular, rackable, workstations resembling the Cray CX1. Unfortunately Apple is moving away from scientific workloads in general.