Comparing 4xSGSSAA vs 4xOGSSAA (downsampling), the downsampling is much more efficient for shading. The higher the triangle density, the more efficient downsampling is compared to SGSSAA. This has to do with how the GPU packs pixel and samples for shading, downsampling packs much better. If anyone is interested I'll go into the details. The only time 4xSGSSAA will be faster than 4xOGSSAA is when for some reason the game's post processing runs at 1x (no MSAA) with 4xSGSSAA (compared to the 4x cost on driver enabled 4xOGSSAA downsampling). Note, some games with built-in in-game 4xOGSSAA might be running post processing at 1x, which will always be faster than driver forced 4xSGSSAA. Also 4xTrSSAA should be faster than 4xOGSSAA, but the shading quality of 4xTrSSAA won't be nearly as good.