Only relatively recently has NVIDIA implemented B-Frames into NVENC. If I am not mistaken, AMD still does not have this capability. I am not deeply well versed

Only relatively recently has NVIDIA implemented B-Frames into NVENC. If I am not mistaken, AMD still does not have this capability. I am not deeply well versed in this space though, but if memory bandwidth is such a huge bottleneck, how does the CPU do it so efficiently comparatively? GPU surely wins in this area? Is it just designed that way so that consumer cards can offer realtime speeds? I'm not sure why this couldn't be configurable in some way.