30 Mar
2012
30 Mar
'12
5:46 p.m.
On 30 Mar 2012, Calvin Morrison spake thusly:
Darrell, David,
Number of cores tends to perform better -j4 for 4 cores. The +1 thing has been disputed numerous times :)
It also depends on hyperthreading, available memory, cache sizes, the cache-friendliness of the compiler (GCC really isn't cache-friendly at all) and a lot of other things.
On my hyperthreaded quad-core 2009-vintage Nehalem (256Kb K2 cache, 8Mb shared L3 cache), I benchmarked GCC 4.5 as taking least time with -j 12 (for a sizeable parallelizable compilation). But it has a hardware RAID array and 24Gb RAM, so it could easily keep up with those demands.
--
NULL && (void)