[Gems-users] Is it normal that threre is no significant difference information collected from warm up checkpoints compare to cold checkpoints

[ruzhu kao <kaoruzhu@xxxxxxxxx>]

Hi all:

    I have simulated STAMP9.10 on GEMS2.1 + simics 2.2.19. When I analysis my results, I found that the breakdown_cycles collected from warm checkpoints are very close to the results collected from cold checkpoints. Is it an expected behavior or indicating that I may do something wrong. I have used ruby0.load-caches to load warm cache data and clear stats. I suspect it is because prefetch  operation in substrate hide the cache miss penalty and make no significant difference. 

  If it is an expected behavior, it seems that cache warm up is not a necessary step in simulation. Is it right?

  Any comments will be welcome. Thanks in advance.

Sincerely yours:

Ruzhu.