>>
So watch your tempdb and create at least one additional tempdb per
socket or even better per core.
>>

Note that this doesn't really at all if you put them all on the same
physical drive.