Cache is king for gaming because of the sheer amount of data needing to be processed. Anything ALU limited in games is categorically not being cache or memory limited because of how memory speed starved modern CPU cores are so making the cores stupid fast only takes you so far. Especially with the insane power budget these chips ask for.

Games need to do a crazy amount of branchy unpredictable work on an enormous dataset so they're very often limited by memory latency which is alleviated most readily by more cache and prefetching. It's why the ECS pattern is so popular as of late from a performance standpoint because it encourages more logic to be written as iterations over dense linear arrays which are just about the best case for a CPUs cache management and prefetching logic.

Factorio is probably the best example of this because it's a game where performance is entirely CPU limited where AAA games more often used in benchmarks are much more dominated by GPU time.

Why is the 7800X3D so much faster in that benchmark than 7950X3D? The 7950X3D seems to boost on paper to even higher frequency. Is the cache different per core?

7950X3D only has the 3D stacked cache on one of the two chiplets, so as far as gaming is concerned you only want to use 8 of the 16 cores. 7800X3D would boost harder for gaming because the 7950X3D has a second vestigial (as far as gaming is concerned) chiplet that eats power that could be better spent in the v-cache enabled die.

Factorio didn't get detected as a game properly so it was half on the vcache cores and half not on it. It's kinda heterogenous compute, and schedulers aren't that smart about it (at least Windows' isn't). If you set affinity to just the 8 cores with vcache then the 7950x3d matches the 7800x3d

When the cache is filled up then they're pretty much tied.

https://m.youtube.com/watch?v=0oALfgsyOg4

Excluding the massive power difference...