Hi, > Sure, scrolling in particular would be better moved to the graphics > card, although I don't see a big problem with it, anyway (even for > the > back buffer case), since you should only update the rectangles that > you > change. > That of course requires the "only draw from the looper's thread" > design > we already have. I want to define scrolling from the accelerants perspective (think textfile for example): -scrolling down: ->blit top-second text line downto last textline one textline up. ->app_server will send new last line using software copying from main mem. -scroling up: ->well, you can guess ;-) It's just using local blits (mainly)! Which makes it so damn fast. And again: no retrace syncing is used. trust me on that ;-) I don't know about non-accelerated drawing though: but it seems unlikely that that would be retrace synced. retrace syncing comes at a huge performance penaltly. (let's say you need 20 times the time of non -synced drawing on high loads) Rudolf.