How do we choose the level of simulated details so that it can realistically evaluate the architecture? If we have too high levels of details, from what I understand, the performance of simulator declines; but if the details are too low, the results are unreliable. It seems to me that it introduces yet another variable that has to be controlled tightly and that adds to some other difficult tasks such as scaling up/down.
bpr
@tommywow, usually an architect will pick several levels of detail. Starting with lots of designs, the faster, lower detail simulations are used to winnow down the possibilities. These designs can then be investigated with higher detail simulations and take into account more constraints and potential interactions with other components that are approaching a settled design. One commonly used simulator, gem5, provides three CPU models for increasing accuracy.
How do we choose the level of simulated details so that it can realistically evaluate the architecture? If we have too high levels of details, from what I understand, the performance of simulator declines; but if the details are too low, the results are unreliable. It seems to me that it introduces yet another variable that has to be controlled tightly and that adds to some other difficult tasks such as scaling up/down.
@tommywow, usually an architect will pick several levels of detail. Starting with lots of designs, the faster, lower detail simulations are used to winnow down the possibilities. These designs can then be investigated with higher detail simulations and take into account more constraints and potential interactions with other components that are approaching a settled design. One commonly used simulator, gem5, provides three CPU models for increasing accuracy.