Apple released a new support document for developers which seems to specify that the only GPUs in upcoming machines will be from Apple.
The give a few reasons for this switch:
This is going to have interesting implications for pro machines, especially desktops. A good overview on the subject here:
If I’m interpreting this correctly, it doesn’t totally eliminate the possibility of external GPUs (eGPUs). But it’s pretty unlikely that Apple will make its own eGPUs and I doubt third-party eGPUs will be readily available or easy to work with.
Or am I wrong?
Got a link to the documentation?
Yes, I deliberately avoided discussing external GPUs.
If you watch the video it is discussed for MBP and iMac, then focused on what this means for the Mac Pro especially.
Screenshots and discussion in this developer session:
Yeah. Interesting implications. Good catch!
I want his t shirt and I like his “whimsy”… 
macOS not supporting supporting nVidia already means no CUDA. So we are stuck with AMD GPU’s and OpenCL. What does “Silicon” mean for GPU processing? Will OpenCL be supported? While Metal is supported by many software makers (Adobe,…) in the graphics/design realm, I am not sure this can be said for number crunching…
I am really curious to see where this will take us…
Old news. NVidia wouldn’t support Metal, Apple froze them out years ago, and they officially dropped support for Macs last year.
Watch the link above.
Being old doesn’t make it better. I have a PC with nVidia just for CUDA applications…
Didn’t give any information on GPU computing. Right now, several frameworks just work on CUDA and not OpenCL. What’s the OpenCL roadmap for “Silicon”?
You responded as if you didn’t know it was old news. 
The roadmap was outlined, whether or not you like it. It will of course get fleshed out in the upcoming weeks, as Apple has done in the past. OpenCL is deprecated, but is available on Apple silicon when targeting the GPU. Google is your friend (if you use it right).
