>> Almost true to say, for architects, you always need to use O3 CPU, instead >> of functional or timing (which are there as a mid-step to develop O3 to my >> understanding). > > There are atomic and timing versions of the simple CPU, useful for fast > forwarding, developing other parts of the simulator, etc. They were not just > to fill the gap while O3 was being developed.
I'd go much further than than to say that I flat out disagree with the statement above. Most architecture experiments involve relative measurements, not absolute measurements. Compare a system with feature X to a system without feature X or to a system with feature Y. There are plenty of cases where a crude CPU model is good enough to get an understanding of relative system performance. For example if you're working on a cache coherence protocol, you can probably find out a lot of useful things by using a simpler CPU model and running for more cycles or running with more cores. Further, if the system that you actually care about uses in-order cores, then error due to the simplistic single CPI machine is even less pronounced. Nate _______________________________________________ gem5-users mailing list [email protected] http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users
