Stefan Mileschin

When Ryzen 5 was first announced we reported that 6 core models would be two CCX modules with one core each disabled (3+3) and 4 core models would be two CCX modules with two cores each disabled (2+2). There was some concern among our readers based on previous reporting that communication between CCX modules has relatively high latency, and that these layouts might result in lower than ideal performance. Hardware Unboxed has quantified what type of differences we'll see between 4+0 and 2+2 layouts. From the looks of it, the theory is correct. Splitting things over two CCX modules instead of keeping it all on the same module does have a performance impact, but it seems pretty small to me.

http://www.hardocp.com/news/2017/03/...ance_22_vs_40/