Hi all, I have been trying to run gem5 with this configuration - 3 level cache L1: 4 cycle, 32kB 4 way L2: 5 cycle, 256kB 8 way L3: 9 cycle, 2MB, 16 way parsec benchmark: vips - on a 4 core machine (checkpointed and run for 250 million cycles) I get the following ipc system.switch_cpus0.ipc_total 0.511700 # IPC: Total IPC of All Threads system.switch_cpus1.ipc_total 0.376984 # IPC: Total IPC of All Threads system.switch_cpus2.ipc_total 0.354908 # IPC: Total IPC of All Threads system.switch_cpus3.ipc_total 0.449460 # IPC: Total IPC of All Threads total: *1.693052*
now when I keep all the parameters same and change the cache latencies alone L1: 4 cycle, 32kB 4 way L2: 11 cycle, 256kB 8 way L3: 30 cycle, 2MB, 16 way I expect the ipc to have decreased even further (given that L2 and L3 latencies have been increased by 6 and 21 cycles respectively) - but this is what I get system.switch_cpus0.ipc_total 1.389916 # IPC: Total IPC of All Threads system.switch_cpus1.ipc_total 1.385153 # IPC: Total IPC of All Threads system.switch_cpus2.ipc_total 1.018119 # IPC: Total IPC of All Threads system.switch_cpus3.ipc_total 1.342527 # IPC: Total IPC of All Threads total: *5.135715* What I observed is core 0 is getting stuck in the 1st setup and it is causing drastic reduction in the number of committedInsts across all the cores. How could the performance decrease (drastically) when I increase the latency? Also I am seeing a decrease in performance (drastically) when I move from 2MB to 4MB for some of the parsec benchmarks. -- *thanks®ards* *BISWABANDAN* http://www.cse.iitm.ac.in/~biswa/ “We might fall down, but we will never lay down. We might not be the best, but we will beat the best! We might not be at the top, but we will rise.”
_______________________________________________ gem5-users mailing list gem5-users@gem5.org http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users
