Continuous LOD research dropped off dramatically when GPUs got vertex shaders. At that point, having the CPU touching the vertex data at all became more expensi

Continuous LOD research dropped off dramatically when GPUs got vertex shaders. At that point, having the CPU touching the vertex data at all became more expensive than it could be worth. Instead, 3 or 4 static LODs have been the way to go. Those static meshes can be individually well optimized. There are surprisingly few good options for auto generating static LODs even today... Mesh Shaders are opening up whole new opportunities that are only starting to be explored. You get far more general purpose, compute shader like features. But, you have to work in small packets. Funny enough, they are a lot like PlayStation 2 vector unit programming made modern. That design really was ahead of it’s time. But, the fundamental physics advantages of SIMD working on a small chunk of SRAM are as valid as ever.