The L1 increased in size as much as it increased in associativity. So aliasing is not gone.isn't that effectively fixed in steamroller?
Going to 3 ways of associativity is better, but still lower than average.
It's better than Bulldozer and Piledriver, but writes are still penalized.isn't that also fixed in steamroller?
An Intel L3 is several times larger, and it services more cores plus a GPU.its also the same size so that kind of makes sense.
It's also not the first thing accessed on an L1 miss.