[RC5] interesting Xeon core anomaly

Nils Petersen nils at psyma.com
Mon Aug 1 04:26:28 EDT 2005


Hi there,

by coincidence (load testing a repaired server) I came across this funny
behavior (2.8 GHz Xeon/HT Prestonia, 496 client core #7 SGP 3-pipe) I
thought I should report:
dnetc -numcpu 1 yields 4,994 kkeys/s (lower than one would expect from
Prescott bench)
dnetc -numcpu 2 yields 2,412 kkeys/s (as expected due to missing SMT
optimization)
but
running dnetc -numcpu 1 and /another/ dnetc -numcpu -multiok
/simultaneously/ yield 2x 2,674,xxx keys/s - not much faster, but
obviously there seems to remain some room for optimization on Xeon.

On my 3.0 GHz Prescott however, this does not work the same way: 1
process/1 thread yields 7,124kkey/s, 2 processes/2 threads yield 2x
3,093 kkeys/s

The Prestonia is reported as 6547:BF29, the Prescott as 6547:0F34.

Best regards
Nils Petersen




More information about the rc5 mailing list