[parsec-users] High CPU utilization for Bodytrack

shervin hajiamini hajamini_2006 at yahoo.com
Tue Jun 3 11:28:00 EDT 2014


Hi all,

I’m running bodytrack on a recent
stable version of Gem5 (with full system mode and detailed CPU). The bodytrack
is supposed to be communication-intensive so the CPUs’ busy utilization (computation rate) should
be very low, however from the stats.txt file I find a high busy utilization
(over 90%) for all the CPUs. I calculate busy utilization for a cpu as (cpu.numCycles
– cpu.idleCycles) /cpu.numCycles and I am using the following command line
options to run bodytrack on 64 simulated CPUs in Gem5:

./build/ALPHA/gem5.opt ./configs/example/fs.py
-n 64 --caches --l2cache --num-l2cache=64 --l1d_size=64kB --l1i_size=64kB
--l2_size=2MB --cpu-clock=2.5GHz --cpu-type detailed --checkpoint-restore=1
--script=configs/boot/bodytrack_64c_simmedium_ckpts.rcS

I have also decreased l1d/l1i
cache sizes up to 8kB but I still get the high utilization for the CPUs. Can
anybody please let me know how I should run bodytrack on Gem5 to get the
expected low utilization for the CPUs?

Thank you very much,
Shervin
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.cs.princeton.edu/pipermail/parsec-users/attachments/20140603/9e615e02/attachment.html>


More information about the parsec-users mailing list