gl_occlusion_culling VS CornerCulling

A low-latency (and frankly better) alternative to hardware occlusion queries is to make your own using indirect drawing. Simply draw a chunk's bounding box against the depth buffer and set the instance count of the indirect struct to one if any fragments are drawn. The next frame you can use these indirect commands to draw the actual chunk geometry and repeat the whole process. This has a guaranteed latency of one frame. More info here.

One approach is to use occlusion queries, draw chunks which were visible last frame and then occlusion test the bounding volumes of chunks which weren't rendered and render the ones which pass using a shader based occlusion approach similar to this NVIDIA sample: https://github.com/nvpro-samples/gl_occlusion_culling

For occlusion culling, specifically, there’s this reference implementation: https://github.com/nvpro-samples/gl_occlusion_culling

Now if this issue can be solved in 3D in a few dozen ms I'm sure in 2D this would not explode the server bill. Also where did you get the 64 tick rate from? Is it from CS:GO because this source says DOTA 2's tick rate is 30. Either way you're overstating the complexity of this issue.