GPU conversion is not as optimal as CPU. Sizes and quality may vary with GPU's depending on the codec. I would say the best server is the one with more cores say a 32-64 core server can convert between 6-10 videos in parallel. I have 32 core server at the moment it does around 100 x 4(qualites) a day. It can do alot more but I dont have more uploads. Just make sure they are not ARM procs and its fine really.
Using video cards requires specific ffmpeg builds and codecs, this means you will need qualified server engineer to build your ffmpeg. Or if providers have ffmpeg pre-built with all the needed nvidia drivers and codecs, this is a good option for video conversion.
I would say not possible to reply this without testing the exact 2 configurations with the same video test. Both parameters are important, CPU clock and the number of cores. Also if you look into modern CPUs, they do not increase clock, but instead they have more performance by increasing the number of cores, and optimizing processing of specific instructions.
Try this article:
There you can find real CPUs and their real encoding times.