Slide 34 of 41
- -Here we see elapsed time plotted for both the dedicated and non-dedicated runs
- -For reference, the single process T90 run completed in ~5 hours
- -Performance gains appear to taper off above 4 processors
- -For the single processor SV1 run, memory bandwidth was our limiting factor
- -Cache utilization remained about the same for all of the runs, so performance benefits were due largely to gains in memory bandwidth.