[Gems-users] L2 latency in different network topologies

[Liqun Cheng <liqun.cheng@xxxxxxxxx>]

Hi there,

I am trying to model a 16-core CMP, which employs a shared L2 cache.
Does L2 miss latency depend on the network topology (how L1s are
connected)? For example, I can use 2 level, fanout 4, hierarchy
topology or 2D torus to connect 16 L1 controllers, I'd like know
whether these 2 configurations have the same L2 miss latency, assuming
the number of banks in L2 are constant.

thanks a lot!
Legion