> I think the most accurate (while still reasonably efficient) way is to detect
> all write accesses to video registers and memory and render everything up to the
> current pixel when one happens. You can even change the trapping range every
> scanline to make things slightly faster. Modern PCs can handle it anyway.

I've also seen emulators that simply store the current scanline each time a write occurs and then play it back like a display list at the end of the frame. It's faster than conventional scanline rendering since you don't need to stop the CPU and context switch, but it can be memory intensive :-)