I am running the PARSEC benchmarks on the Alpha Full System mode with 16 cores. I have ran various simulations where the only setting changed was the functional unit pool for a single cpu out of the 16.
It seems that even when I use a "slower" functional unit pool, sometimes the benchmarks complete in LESS time than when using a "faster" functional unit pool. I have used diff on the config files and the only difference is the functional units. It seems as though the work gets split among the cpus inconsistantly when checking the stats files, so trying to get meaningful data using various functional unit pools as the only setting being changed is impossible. Is there a way to have the Alpha FS divide the PARSEC benchmark work among the 16 cores the same way every time? Thanks. -Matthew Catanzaro _______________________________________________ gem5-users mailing list gem5-users@gem5.org http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users
