I grabbed this multi window test code, changed it to use D3DXCreateTeapot
instead of D3DXLoadMeshFromX
(I couldn't find a teapot.x file), moved the EndScene
call below the DrawText
call and set NUM_WINDOWS
to 1. With those minor changes, the test works and creates two windows, each with its teapot.
I built the test and deployed it in a machine which has an Intel HD Graphics onboard GPU with two heads, each attached to a monitor. Then I moved one window to each monitor, and enlarged both windows to take up about 80% of each monitor space.
With this setup, which is quite close to what my app needs, the window in the secondary monitor always goes too slow. If I swap the windows, it's the same: the one in the secondary monitor starts crawling, and slows down the whole system.
I googled around and some sources (albeit dated) state that only the primary monitor can use hardware acceleration when not in fullscreen mode. I cannot use fullscreen because direct3d9 rendering in my app is done inside a user control embedded in a Winforms GUI.
Is it really impossible to get hardware acceleration for both monitors in windowed mode? The legacy version of our application uses MFC + DirectDraw and manages to perform fast enough, but those are obsolete technologies and we'd abhor going back there.