Hi Chris. Assumption 1 is incorrect.
Resolve uses CPU for codec transcoding and GPU for all image processing. So files from disk are decoded, if required, passed to the GPU for processing. If they are RAW they are debayered in GPU. The exception to this rule is if you have a Red Rocket as the decompression and debayer is made in the Rocket, if not Rocket the r3d files are decompressed and debayered in CPU as thats what the RED SDK requires. In the render process the graded images are then passed from GPU’s to CPU to package into the codec of your choice, assuming you compress, and then to disk.
So to speed your renders, you need to optimize a number of processes, or at least find the speed bump. It could be the disk speed, raid controller, number of CPU cores for decoding, decompressing, debayer (if r3d). Then the number of nodes, NR, Blur, and GPUs used to process the images. The resolution, image resizing etc, all uses GPU resources. Then back to the CPUs for compressing, the Raid controllers ability to switch from read to write dynamically and then disk speed again.
My guess based on your details above, by setting the Red Rocket for a Full Res Premium decode of the r3d, the RED Rocket is making a 4K image, which then your single Q4000 GPU must resize to HD for the timeline before it applies the corrections. Try a 1/2 res setting on the Rocket and see what difference that makes as this will give more GPU power to be used for the grades instead of resizing.
And yes, 100+ fps render is very possible with a multiple Rocket, GPU system on a fast disk array.
Peter