Hi all, I have a question with my running result of facesim and streamcluster application of PARSEC. I tested them on different cache configuration with the same 8-core CPU config. my cache configuration is: same on L1 cache: 2-way, 64 bytes block size, 64kb IL1 cache and 64kb DL1 cache
different on L2 cache and L2 cache is the last level cache. a). 4-way, 8M size of L2 cache b). 8-way, 8M size of L2 cache c). 16-way, 8M size of L2 cache Normally, the higher way set-associative L2 cache config will have less misses, but my experiment results show the 4-way cache has fewest misses, 8-way second, and 16-way has the biggest misses. However these only happen to facesim and streamcluster, other applications are reasonable. So i want to know whether any one come across with the same problem or something wrong? Thank you in advance. Sheng
_______________________________________________ m5-users mailing list [email protected] http://m5sim.org/cgi-bin/mailman/listinfo/m5-users
