Just updated my 5 year Mac to a PC, i7 7800x with a single NVIDIA ROG Strix GeForce GTX 1080 TI OC.
For a given scene rendering for Print (5000 x 5000px) in cycles and with high poly and hair particles...
GPU rendering was marginally faster than my late 2012 Mac (4 threads).
CPU rendering (12 threads) was significantly faster at over three times the speed.
Results: (approx. values)
Mac render 1 hour & 45 minutes.
GPU render 1 hour 20 minutes.
CPU rendering 35 minutes.
- This was a single render test.
- There is a chance the PC and/or GPU was not optimised for Blender
In this test though the factor was thread count.
There is also a chance that SLI could significantly speed up GPU renders to compete with the CPU.
In my experience the GPU is optimised for calculating and displaying particular graphics quickly ie. low poly, normal mapping, simple particle arrays, physics, shadows (which is always hard to do in real time), shading - post processing effects. Most effects we see in games such as caustics are usually faked using shading or texturing and not often calculated in real time.
For the price of an Intel Core i9 7900X, which is comparable to the price of my GPU I could have got another 8 threads at 20 threads, and use the money I spent on the 7800x to buy a lesser GPU.