If that's a reference to the recommendation that I mention sometimes, based on a conversation with Adam Machanic, then you've got the formula the wrong way round. It's not 1.5*(number of NUMA nodes) as that would be quite weird and highly unlikely to produce an optimal config. It's 1.5*(Number of physical cores in a NUMA node), so 12 (assuming that hyperthreading is enabled)

He hasn't, to my knowledge, published his testing and, based on what I know about the kind of systems he works with, is probably for massive anaytics-type systems, and so I would recommend either 8 or 12 depending on whether the system is more OLTP or more OLAP.