I am simulating a 4x4 Mesh_XY (16 nodes) with neighbor traffic pattern and following parameters:
--cacheline_size=512 --sim-cycles=1000000 --sys-clock=1GHz --injectionrate=1.0 --ruby-clock=1GHz --inj-vnet=2 --routing-algorithm=1 --link-width-bits=256 --link-latency=1 --router-latency=1 --vcs-per-vnet=4 --buffers-per-data-vc=3 Also, I set the clock domain of links and routers to 1 GHz. With neighbor traffic pattern and 1.0 as injection rate, I assume each node is sending a flit to its neighbor in each cycle. Thus, in a steady-state where the network pipeline is filled, each node should receive a flit from its neighbor. Therefore, we should have an aggregate bandwidth around this number 16x256x10^9 bit per second(#nodes*link-width*clock)=119.2GB/s. However, the total received flits reported by Garnet is 1984 flits for 10^6 simulation cycles(10^-6 seconds) which lead to 1984x256x10^6 bit per second=59.12 GB/s bandwidth. Does someone have an idea why there is a large gap between the peak bandwidth and reported bandwidth in Garnet? _______________________________________________ gem5-users mailing list -- gem5-users@gem5.org To unsubscribe send an email to gem5-users-le...@gem5.org %(web_page_url)slistinfo%(cgiext)s/%(_internal_name)s