If you look at the video generation circuit on an Apple ][ it's all done in real-time in TTL, alongside the DRAM refresh activity. And it's all crammed into the normally unused tock of the 6502, making this work with virtually zero overhead on the CPU. That's the reason why the memory mapping of the screen isn't linear to the physical display. The software tradeoff was cheaper than the hardware one.
http://twimgs.com/informationweek/byte/archive/Apple-II-Desc...