Personally I get this kind of "this device really, really needs DirectDraw now" feeling..
![Smile :)](./images/smilies/icon_smile.gif)
I guess one could try some reads/write to the memory address returned from GXBeginDraw and compare the performance from plain memory. I never remember if it was reading or writing that didn´t use cache on the ARM (I guess it was writing), but it could show a difference.
My colleagues are playing with the 740 right now, and I will dig some more when I get back. I guess an email to ATI wouldn´t hurt either (I mean, all libraries out there would benefit from some kind of API, even a low level one), so I'll send one that way as well...
/Johan