https://hgpu.org/?p=1265
Evolution of thread-level parallelism in desktop applications