num is the number of data units processed by a single instruction, lat is the instruction latency (the number of cycles after which the result is ready) and thru is the instruction throughput (the number of cycles after which another instruction can be issued).

SSE	num	lat	thru	3DNow!	num	lat	thru
addps subps maxps minps cmpps	4	4	2	pfadd pfsub* pfmin pfmax pfcmp*	2	4	1
addss subss maxss minss cmpss	1	3	1	-
mulps	4	5	2	pfmul	2	4	1
mulss	1	4	1	-
rcpps rcqrtps	4	2	2	pfrcp pfsqrt	2	4	1
MMX-mult		3	1	MMX-mult		4	1
other MMX		1	1	other MMX		2	1

Note that practically the result of a 3DNow! instruction can be fetched by another instruction after 3 cycles, even though it is ready after 4 cycles.