Hmmm... I think you should not worry too much about matching stats with the paper because the ROI may be different from what you are using to measure..it's enough if you see the required activity on the frame buffer..Also these are multi threaded apps..doubling the number of cores does make a difference.
Srini <div>-------- Original message --------</div><div>From: Vishaal Mohan via gem5-users <[email protected]> </div><div>Date:06/04/2014 1:43 AM (GMT-06:00) </div><div>To: gem5-users <[email protected]> </div><div>Subject: [gem5-users] Moby (Asimbench) statistics mismatch </div><div> </div>I've been trying to run a few benchmarks from the Moby benchmark suite. The results that I'm getting are no close to the ones reported in the paper (asg.ict.ac.cn/projects/moby/downloads/Moby-ISPASS2014.pdf). The parameters that I was using are slightly different but I still think that the difference is too much. Configuration I'm using: L1 icache and dcache: 4 way associative, size 32kB L2: 16 way associative, size 512kB Memory: type-lpddr2_1066_x32 size:256MB(The size given in the paper is 1GB) cpu type: arm_detailed number of cpus: 1(The paper uses 2 cores) I havent made any other changes to the source files, like access times. For example, for the mxplayer benchmark the icache overall miss rate is more than an order less than that for dcache. In the paper, it is given the other way around. Is this because of the difference in the parameters that I've taken or have the authors made other changes? Thanks a lot. Cheers. --Vishaal Mohan
_______________________________________________ gem5-users mailing list [email protected] http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users
