Considering the iMac Pro was meant to be the top-end Mac and on that machine the GPU is not really upgradeable (it is soldered to the system board) and considering the GPU is the one component that sees the most frequent development, if any component on the Mac Pro will be upgradeable, common sense says it would be the GPU.
the premise there is twisted. There is not much indication that the iMac Pro was meant to be the end all permanent capstone top-end Mac. Its primary purpose is to be more "horsepower" than any iMac has ever had. It happened to race past the current Mac Pros, but that isn't particularly hard since the MP is comatose with 2013 technology and the iMac Pro has 2017 technology. Yeah, the iMac 4 years later should be catching up to at least some of the Mac Pros from
years gone by. That is more than two Moore's law cycles ( ~18 months x 2 )
The primary reason the iMac Pro is on top now is that the Mac Pro is comatose.
The iMac Pro's GPU is good enough for very large number of folks. For more than a decent number it is overkill that will last for their workloads for 3-7 years until they cycle through to another machine.
The second issue is the "GPU most frequent". Really? Lots of stuff gets lost in the Nvidia vs AMD fanboy wars. Nvidia's Pascal came out in 2016 and hasn't really been replaced since. That is about 2 years (something new is coming within a year). Nvidia isn't particularly doing any better than Intel is with the "Xeon E5"/"Xeon W" class CPUs in terms of pace. They milk the architecture roll out longer than Intel does but by and large they have the same slowing fabrication process technology limitations affecting everyone else in the "big die processor" world. ( Volta is being rolled out to Quadro and probably into Titan eventually but same baseline microarchitecture that will be about two years old at that point when they finish. )
AMD has been behind the curve relative to both of those vendors ( although it seems like they have shooting themselves in the foot at this point. ). Vega hasn't effectively filled Polaris slot. Again it is activity by dropping different ranges in different years and doing a couple of rebadges. It is sales pitch/price movement but not major arch/performance moves.
GPU's pace hasn't been better. Top end GPUs thirst for higher power limits (TDP) has been what has been on a lust curve. Intel has normalized where the workstation CPU TDP is. The GPU folks have been on a growth curve in terms of max power levels.
And Apple has been happy to allow third parties to offer upgrades to CPUs, GPUs and SSDs for the Mac Pro so common sense says that they will allow third-party GPU updates even if they also offer their own "official" ones.
People have done CPU upgrades but not sure Apple has really be happy about that (at least while still covered by AppleCare and/or warranties). Apple has had their own SSD path for a while in new Mac systems.
I think what is more commonsense is that Apple needs to allow some room for 2nd and 3rd SSDs and GPUs that aren't on Apple's specific path. The CPUs aren't really an issue. Most of that is really when the systems are 3+ years old when the price for used CPUs starts coming down and bargain hunters get more active.
[doublepost=1522434553][/doublepost]
They released the trashcan Mac Pro and never referred to the lack of upgradability as an issue. They only pointed out the thermal problems.
They seem to favour eGPU. I would be surprised if the next Mac Pro came with PCIe slots.
They did point out "high bandwidth" as being an issue they are missing (with MP 2013 and defacto missed by the iMac Pro). You can't cover x8-x16 bandwidth with TBv3. So eGPU isn't a panacea. Saying "high bandwidth" is covered by eGPU enclosures is just being willfully ignorant. They'd be walking away from RAW 8K video capture. They'd be putting a throttle on their GPU solutions. Games will play OK (better than the embedded GPU) , but high data throughput GPGPU workloads won't.
iMac Pro discards about 20 PCI-e v3 lanes of throughput from the CPU. Not sure who you tackle the "high bandwidth" needs market throwing away that much. eGPU are a poor substitute for that. ( going to a grotesque number of TB ports, 6+ , isn't really effective either in cost or bandwidth).
