That was 1024^2, just to get nice quality. However 512^2 and 3 slices looks pretty darn good with even just 4x MSAA on the shadow map (G80 supports MSAA for fp32 render targets!)... even 2 slices of 512^2 looks pretty nice IMHO, although a bit less detailed naturally. Probably 512^2 with 3 slices is the best trade-off right now on the G80, although it will of course depend on the scene.

I'll be posting the demo binary publicly soon so anyone with a G80 and Vista (D3D10) is welcome to play around with it: pretty much every option is exposed in the UI.

Just for the record, 672 is a weird size ;)

Quote:Original post by wolf
With 4xMSAA my 672 maps are like 1024 maps ... pretty cool.

Yeah multisampling with VSMs is awesome - it makes static shots look better, but it *really* makes a huge difference in motion, completely eliminating swimming edges in many cases.

Anyways I'll be posting the demo soon... just have to clean up a few things :)

Mine's still not really working. Here's my SampleState and my horozontal blur. It works fine on the first slice, but the 2nd slice is pixellated like a traditional shadow map, and beyond that, everything is in shadow.

SamplerState Linear
{
AddressU = Clamp;
AddressV = Clamp;
Filter = MIN_MAG_MIP_LINEAR;
};

float4 PS_BlurH(PS_Blur_In input) : SV_TARGET
{
int softness = 3;
float4 output = (float4)0;
float step = 8.0f / (512.0f * BlurWidth[LightID]);

float2 tex;
for(int i = -softness; i <= softness; i++)
{
tex = input.Tex + float2(i * step, 0);
output += ShadowMap[0].Sample(Linear, tex);
}

return output / (2 * softness + 1);
}

I tried SampleLevel, and it produced the same results. The softness is actually like a 7x7 box blur, just clearer code to me, and the BlurWidth is the width of the corrosponding orthographic projection matrix.

And what are the code tags on this forum?