Re: new d3d9/device.ok test always fails here, but not a regression?