Man, that design sure didn’t age well. Will be interesting to see how Apple reimagines the new Mac Pro form factor, but I think we can safely say that we will never be seeing a rehash of the old tower design.
Just for kicks, I went back and re-read Anand Lal Shimpi's review of the 2013 MacPro -
https://www.anandtech.com/show/7603/mac-pro-review-late-2013
The review is exhaustive and once I read it, I understood better the compromises that Apple's engineers were forced to make given the parameters with which they had to work within. Unfortunately, the parameters are largely self-inflicted - with one MAJOR exception. As we approach 2019, my hope is that same mistakes are not repeated.
While the Xeon CPUs Apple chose to use were fine, the C602J PCH is/was
the Achilles Heel of the 2013 MacPro and would have been even it Apple had opted to create a traditional cheesegrater tower (more on that later). All of the C6xx PCHs of that time were based on PCIe 2.0, and contained a meager x8 lanes of
PCIe 2.0. Adding insult to injury, the bus connecting the 4-core, 6-core and 8-core CPUs to the PCH was/is a DMI 2.0 bus (5GT/s), maxing out at 2GB/s. Looking at the diagram, the amount of ports hanging off of that PCH is both amazing AND disheartening, once the compromises come into play.
The review reminds us that the C602J PCH did not natively support USB 3.0 and Apple had to add support using a single x1 PCIe lane attached to a Fresco Logic USB 3 controller. With the benefit of hindsight and my recollection that Mac users back in the day (2011-2012) were at a fever pitch for Apple to add USB 3.0 support to the Mac lineup, I am pretty certain the average Mac Pro buyer would have been fairly seething to be asked to purchase a Mac Pro in 2013 without USB 3.0, considering that every other Mac being sold at that time had at least two (2) USB 3.0 ports.
I am also betting that the same users would have also been seething had Apple neglected to include SSD storage as the default storage choice in the 2013 Mac Pro, although some will argue that a 2.5" SSD would have been good enough for them at the time. To that I say, "Mule Fritters!!!".
Had Apple gone the traditional tower route, they would have been somewhat better off, but not completely out of the woods. The constraints on the USB 3.0 ports and the SSD storage would still remain, as neither of those could/should be hung off of the CPU back in 2013.
This brings me to the CPU and the x16 lanes that Apple used for the second GPU. At the time of the Mac Pro's development (2011-2013), IIRC, multiple GPU rendering was viewed as viable and the next big thing. One wonders if a PCIe 3.0 switch would have been possible after reading this article -
https://www.microway.com/hpc-tech-tips/common-pci-express-myths-gpu-computing/ - although I do not know if this was a viable path back in 2011-2013. Also, with a single GPU card in the x16 PCIe 3.0 slot, it really would not have mattered that much anyways.
Freeing up those x16 lanes would have allowed Apple to give us one x8 PCIe 3.0 slot and two x4 PCIe 3.0 slots on the Mac Pro's motherboard, along with the six (6) Thunderbolt 2 ports. That would have been pretty tantalizing to most Pro users who were fairly underwhelmed after the introduction of the 2013 Mac Pro.
The biggest downside to this scenario, and the one I believe was the main constraint on Apple's engineers, was what to do with those six (6) Thunderbolt 2 ports. At that time, Apple was determined to make Thunderbolt a thing for passing DisplayPort signals as they were (and still are) seemingly obsessed with the single port, do everything approach. This obsession goes all the way back to the Quadra 660AV and 840AV (DB15) and the Apple Display Connector (ADC). The lack of an elegant solution to allowing a user to attach their Thunderbolt display to the Thunderbolt 2 port on the back of their Mac Pro while still being able to tap the power of the discrete GPU plugged into the x16 PCIe 3.0 slot, whatever the engineering issues were back in the day, I believe are what led Apple to move to the non-expandable cylinder still currently on sale five (5) years on.
I will be very interesting to see if Apple has solved, or even tried to solve, this particular issue as it relates to allowing users to exchange GPUs when faster ones become available in the future.
SPOILER ALERT: I do not think they have, nor are they trying to engineer a solution. I think Apple will sell modules with fixed GPUs engineered into them and/or license it to third parties (maybe). Again, I do not think the Mac Pro will use cables to connect these modules together, but will use some sort of proprietary connector with a proprietary interconnect standard and a backplane if they have figured out how to pass PCIe 3.0 over/through/encapsulated within PCIe 4.0 to allow full bandwidth and first class citizenship to each and every module within the limitation of the CPU and PCH PCIe lane counts (~x68 lanes total for 1S configurations). The next year should be very interesting indeed!!! And possible maddening as well!!! Happy New Year!
.
