If the Primitive shader is only taking advantage of the instruction set available to the CU and nothing else, then this doesn't make sense to me because you might as well run your shader code from your Compute shader, and test for a performance path output to the FF rasteriser when performance is required from them. Compute shader or Primitive Primitive shader, they both have access to CU instructions. Unless your Primitive shsder has access to additional instructions from a set of ALUs which may also include CUs, then these additional instructions made available to your Primitive shader makes switching worthwhile.Primitive shader most likely takes advantage of the ALU in the CUs.
Anothet thing that doesn't make sense to me is that if both Epic and Sony were in discussions about future rendering technologies, and Cerny had access to AMDs IP portfolios, why didn't they design an advanced hardware Programmable Primitive block that would act as a fast function 'Reyes Rasteriser' processing micropolygons - similar to how traditional FF rasterisers act today but designed for REYES? This should leave the compute heavy work to all your CUs to focus on lighting and shading for Lumen.
Tim Sweeney has been thinking about REYES and voxels since at least as early as 1999. Below are his predictions from then for circa 2006/7:
https://techreport.com/news/46/sweeney-on-the-future-of-3d-graphics/
"2006-7: CPU’s become so fast and powerful that 3D hardware will be only marginally benfical for rendering relative to the limits of the human visual system, therefore 3D chips will likely be deemed a waste of silicon (and more expensive bus plumbing), so the world will transition back to software-driven rendering. And, at this point, there will be a new renaissance in non-traditional architectures such as voxel rendering and REYES-style microfacets, enabled by the generality of CPU’s driving the rendering process. If this is a case, then the 3D hardware revolution sparked by 3dfx in 1997 will prove to only be a 10-year hiatus from the natural evolution of CPU-driven rendering."
I can excuse his CPU to be modern day Compute in GPUs today. His REYES became Nanite, and voxels became Lumen. I suppose RDNA2 will be like a crossroads, allowing traditional rendering to exist with emerging alternatives.