They very specifically said it was an interconnect issue. With the extra L3, adjacent dies made so many requests to each other's cache that it turned into a bot

They very specifically said it was an interconnect issue. With the extra L3, adjacent dies made so many requests to each other's cache that it turned into a bottleneck, and this wasnt an issue on Milan X because its IO die is apparently beefier. I cant find the source because I am on mobile and now google search is flooded with X3D chatter :/