I’ve tidied up the test so anyone should be able to run it.
HitTheWallTest.cs (3.5 KB)
Here’s the plotted result
It’s linear up to hitting the wall. I’ve tested with memclear and no memclear. Makes no difference.
I’m testing this on AMD Ryzen 5900X. I’m really interested how Intel handles this.
The wall start appearing around the limits of L3 cache. Although, it says MB in the chart. The MBs in the test get increased by the NativeList.Resize to a power of 2 so it’s not accurate and as 2 lists are used, the amount of list bytes are actually double.
Still, it aligns quite well for the 64MB L3 cache I have.