I’ve been working on implementing a Forward+ tiled lighting system for a student project. My initial code was based on AMD’s TiledLighting11 demo from their site, and works well on my team’s AMD/ATI machines, however, when we move it to the NVidia machines we have for testing, it can’t seem to access buffer memory over around the 12Mb mark.
The buffer in question is a RWStructuredBuffer we’re accessing via a UAV in a compute shader and a SRV in the pixel shader to generate and use index lists of lights. Any attempts to access the buffer past around the 12Mb or so mark simply return 0 in the pixel shader. This problem doesn’t arise on the ATI machines we have.
I was wondering if there is perhaps a max buffer size/access constraint that is being imposed by the NVidia systems that isn’t present on the AMD ones for some reason?