I think it was a system mem buffer -- copy the data from vram to ram all at once, THEN mess with it.
@Frank: how fine a grain are you scanning -- I'm doing 20-30 samples a minute at 640x480 without any special optimiation. With the OR'ing of adjacent cell data, I would think you could stay fairly coarse without losing that much.
If you would like -- I can poke at the network code for you. Shouldnt be difficult to have a master assign coords to slaves, and the slave do the full 360 scan and return a visibility set. Are you streaming the models across the link or using a shared file system ??
@Frank: how fine a grain are you scanning -- I'm doing 20-30 samples a minute at 640x480 without any special optimiation. With the OR'ing of adjacent cell data, I would think you could stay fairly coarse without losing that much.
If you would like -- I can poke at the network code for you. Shouldnt be difficult to have a master assign coords to slaves, and the slave do the full 360 scan and return a visibility set. Are you streaming the models across the link or using a shared file system ??