The DCT methodology collects a number of performance counters for each application phase in a limited number of configurations (consisting of thread counts at various hierarchical levels, e.g., active cores and SMT threads per core), and uses a hardware-specific predic- tion model to extrapolate application performance for each configuration.