Here's my lame-o 4K composite, maybe like 35 minutes rendering time (all CPU, no Nvidia CUDA), just to show how I decompose this shot... which is a lame shot because I spent 0 time on the background, and I could do more, but I just threw this together after I set up the pose for another shot.
Background is a separate layer series, shot at lower resolution and rendered in like 5 minutes while I was doing other stuff. I don't care about that because it's going to be blurred in a simulated DOF. So who cares? The trick was lining up the HDRI with the general direction of the lighting in the main image, but that's fast because it's a stripped down scene at lower res.
The main image is made up of my 2 figures. I let that run the longest, simulated at higher than 4K resolution (while I was working out). Since it's not full width, I don't care about true 4K here, but I DO care about pixel density where the figures actually are.
After that, it's just a composite shot (with some focus/lighting effects), which I finally exported to 4K UHD.
So, like 50 minutes from concept to final asset.