[data from a dual quad-core 2.7 GHz Opteron 8384 (Shanghai)]

PERFCTR INIT: vendor 2, family 16, model 4, stepping 2, clock 2699957 kHz
PERFCTR INIT: NITER == 64
PERFCTR INIT: loop overhead is 856 cycles
PERFCTR INIT: rdtsc cost is 170.9 cycles (11799 total)
PERFCTR INIT: rdpmc cost is 21.7 cycles (2245 total)
PERFCTR INIT: rdmsr (counter) cost is 174.6 cycles (12034 total)
PERFCTR INIT: rdmsr (evntsel) cost is 198.3 cycles (13548 total)
PERFCTR INIT: wrmsr (counter) cost is 252.6 cycles (17027 total)
PERFCTR INIT: wrmsr (evntsel) cost is 779.1 cycles (50724 total)
PERFCTR INIT: read cr4 cost is 36.7 cycles (3209 total)
PERFCTR INIT: write cr4 cost is 225.4 cycles (15285 total)
PERFCTR INIT: write LVTPC cost is 70.7 cycles (5387 total)
PERFCTR INIT: sync_core cost is 395.6 cycles (26180 total)
perfctr: driver 2.6.36, cpu type AMD K7/K8/Fam10h at 2699957 kHz