What I'm thinking of wouldn't really be a Multi-Chip Module. I'm think of more how Ryzen has two CCXs on the same die.
[...]
Reading the patent application for Split Frame Rendering "The Processor includes a central processing unit (CPR), a graphics processing unit (GPU), a CPU and GPU located on the same die, or one or more processor cores, wherein each processor core may be a CPU or a GPU."
You mean a more rigid slicing of the units and performance degradation for the sake of chip complexity?
Thats like really a cost optimization, where I am due to the parallel nature of GPUs and better scaling mechanism in comparison to CPUs wouldn't welcome such grouping.
Edit: Well it does help to just google the paper and read it. :)
http://www.freepatentsonline.com/20180211435.pdf
By skimming over it, it looks pretty much like the slicing AMD currently does but with better work distrubution.
As far as I understand it AMD currently slices up their GPUs to 1-4 pieces which AMD calls a Shader-Engine, each SE has one Geometry/Rasterizer-Engine which are connected through a crossbar and can be synchronized but work distribution seem to be rather stiff or was pretty bad in the past for AMD, where Nvidias Front-End is/was much better.
After the Fixed-Function-Frontend are the Compute Units, the ROPs and L2$ tiles connected to corresponding memory channels.
The improved approach seems to have multiple Command Processor who can feed the Shader-Engines with 3D or Compute-Queues, currently there is only one Graphics Command Processor feeding all of them with 3D or Compute-Queues, there should be another Processor which can only handle 3D-Queues.
Since GCN2 there are two compute processors which have up to 4 compute pipes and feed all SEs/CUs with Compute-Jobs.
The SEs should have better mechanism to devide and distrubute geometry/rasterization-work between each other.