saw some of you talk abt the ryzen memory problem... what exactly is it?
Basically if you throw it a piece of code that isn't specifically optimized of Ryzen dual-CCX architecture (e.g. 99.99% of all the amateur programs out there and a good % of professional codes) the L3 cache goes "hurr durr" and people cry out "low performance"
Easier to explain with die shots. The below one is of Skylake.
And this one is for Ryzen.
Note how the L3 cache is distributed. Basically it is slower if it needs to get data from the other group of cache.
Of course, why nobody ever complained about this on multi-die / multi-cpu systems like since multi-die / multi-cpu ever existed, is a big fat mystery, like somehow when this happens within die level instead of outside then "omg the program is not optimized I'm unfairly suffering lower performance".
Oh wait, I remember why. Because people accepted the fact that each CPU or GPU core only gets guaranteed performance from their own dedicated cache amount. Oh look at the datasheet it says 2 x 8MB right there.
And also, the Windows scheduler patch that people have been asking for has already been released last month and nobody seemed to have given a shyt nor written an article about it. What's interesting is that AMD has already said it isn't Windows at fault but people don't listen.
In short, if anybody else mentions the cache issue, show them this comic.