The biggest change for me was L3. The memory latency was something in the VM itself and some tweaks.(I purged the whole VM) The L3 was a 6x decrease in latency. L1 for me was 3ns now 1ns and thats massive for L1 as its used non stop but the biggest thing is that the cache is properly allocated. Before if you look i had 5x16 on my L3 which is literally impossible and L1 was 2x larger than it was supposed to be and only 2 way instead of 8 way.   Also with this change you can't span numa.