You mean Metal 2/3 long live. The RDNA Architecture has nothing to do with the new Mac Pro, presently. The base model has the GCN Polaris 580 and then it's Vega Pro. It's HBM2 behemoths are GCN Vega VII Pros. The upcoming RDNA II is meant to be the first big Compute capable designs to compete with Nvidia 2080Ti and Lisa Su has said both HBM and GDDR6 based versions will appear.
The 5700 (RDNA / Navi ) is on the
list of eGPU supported cards. There is are extremely good chance that will work in the new Mac Pro also. The Mac Pro won't launch with a ( RDNA / Navi ) MPX module, but the new Mac Pro isn't solely limited to MPX modules.
Nvidia 2080 isn't "big compute". Nvidia has V100 and T4 products for "big compute". The big RTX (TU-102 ) dies through lots of die space at niched raytracing and tensor which is only "big compute" in a very narrow area. RDNA II is going down that same niche, but it AMD appears to have something else aimed at more general "big compute " computation.
https://www.tomshardware.com/news/amd-arcturus-vega-gpu-support,39935.html
https://www.tweaktown.com/news/68074/amds-next-gen-arcturus-gpu-teased-here-1h-2020/index.html
There is a "bigger" RDNA coming but it is not clear that is lined up with what Apple wants. ( fast track Macs for top end gaming market? Probably not. ). For 7nm (and smaller) going to have to make more tradeoffs at the max die size gets smaller. Can certainly go bigger than what Navi currently is, but will be less able to throw everything and the kitchen sink into one every expanding die size.
The supposed "Nvidia killer" that AMD is suppose to have in RDNA II is more narrowly focused than "big compute". It is probably more so on the gaming side ( as some will be folded into the upcoming 2020 gaming consoles too. ).
If there is a "mid-rangel" RDNA II solution in late 2020 then perhaps Apple would do a MPX 580X replacement with that instead of using something like a 5700 baseline. But "big card" it more likely looks like Apple would take Arcturus (with some HBM bump) than RDNA II.
There is a reason Radeon HBM2 based cards were limited run on the Radeon VII--both Samsung and SK-Hynix have focused on the successor--HBM2e.
Probably a bigger reason was that the Radeon VII was a either a money loser or very close to break even. Those were probably allocated as offsets to the MI50 and MI60 they sold are much higher margins. Using a "big" die as a pipe cleaner for getting onto the 7nm process was probably more than a little expensive.
It also wouldn't be very surprising if the Vega II was late ( which will probably have a substantively higher price point) and so AMD also have 'extra' dies they needed a 'sink' to divert some supply to. And on top of that Navi was also also substantively late ( had to do a bug fix tape out for fixes. )
I'm sure AMD will use HDM2e (epsecially in some upcoming Apple products ), but that wasn't the major hold up on Vega20 microarchtiecture.
Apple needs as much Metal capable resources it can get its hands on, from Macbooks to Mac Pros. CUDA is a competitor to Apple Metal APIs--something nearly everyone on this site overlooks. Apple has zero interest nor need to add Nvidia to its mix of AMD discrete and Intel integrated GPU capable solutions for Metal. Apple took the best of OpenCL--their invention, and released Metal long before Vulkan/DX12 were a reality.
Errr. not really. Especially in a macOS context.
Metal on macOS. 2015
Mantle ( pre-cursor foundation for Vulkan ) 2013 (
https://www.anandtech.com/show/7371/understanding-amds-mantle-a-lowlevel-graphics-api-for-gcn )
Vulkan 2016 ( but started off when AMD handed over Mantle in 2014-2015 as 'starting point' because wide consensus was extremely slow in developing around what "gl-next" would be. Mantle was a far more concrete starting point than "green field" committee product. )
The pre-cursor to DX12 was also in the Mantle timeframe.
There was lots of research into what were the choke points on OpenCL in the 2009-2012 timeframe that had several groups moving toward very similar solutions. Microsoft's , Nvidia , and Googles somewhat roadblocking OpenCL also occurred around that time. ( and OpenCL 2.0 taking some specs (more C++ flavor) that Apple didn't entirely like. ). If there wasn't as much committee 'drama' around 'gl-next' and 'OpenCL' Apple might have gone that way. Instead what they were seeing is a couple of powerful players trying to push a more fragmented foundation layer. Frankly, Apple had the money and resources to go that route too. Coupled to the iOS growth path they'd have a very large ecosystem (and userbase ) to drive demand too.
Metal was initially only primarily a shading language replacement. Its general purpose compute was (and still is a decent amount) relatively limited. Apple's tossing OpenCL for Metal isn't a good fit as much as it is convenient (for Apple ) .