Well, AVX512 would not help that much. As far as I know there is almost no calculation that can benefit from that in games. And if used, the cpu-cores clockrate is decreased quite significant.
It's probably the most GPU-like structure on a CPU, as a massive SIMD array. Games don't use it right now, because they're not trying to run graphics on the CPU like this video, but I mean if they were reaarchitected completely to take advantage of a CPU, that would be an interesting area to look at for the most GPU-like parallelism. You'd have to drop clocks just like you have to for multicore turbo, but the aggregate performance is still higher.
It was more or less a direct response to GPUs gobbling up more datacenter BoM, after all.