Try uploading a very low resolution fog of war masks and then rendering that using a pixel shader. Calculating a huge fog of war texture and uploading it per frame can't be good for performance

Can't you just render the fog of war using spotlights? The cones equal to the light, and everything is shaded dark unless it is in a cone (cap value for multiple cones could be necessary). Using deferred rendering with hlsl shaders should be quite fast, in combination with a quick check if a cone is on the screen. So this would work gpu side only. If you need the fog of war data on the cpu side (for example : visibility determination of units) you could just calculate the fog of war's boundings at a low resolution..

Well you could have a rendertarget texture where you write a 0 to if a pixel has once been lit by a cone, 1 is default. If you use deferred rendering, this texture would simply have the size of the screen. Now on the cones lighting, you read out the value of this texture and multiply it with the lighting component (1 means that a pixel gets lit, 0 means no more lighting). You now clear the texture to 1 once a frame. You get the idea? I don't know if you want to use deferred rendering, but the same idea would work as a postprocessing effect. You could use a red color only float texture as rendertarget, for performance purposes. If I was to make a strategy game I would do it that way.