The superlinear increase in simulation time as a function of system size is a major challenge for computer architects in academia and industry. In this project, we propose computer architecture scale models: a novel methodology to predict the performance for large-scale general-purpose computer systems, both CPUs and GPUs. Scale models are scaled-down versions of the target system of interest. The key benefit of a scale model is that it is feasible to simulate using existing infrastructure. Extrapolation models are then employed to predict performance for the large-scale systems that cannot be simulated (or are impractical to simulate) using existing technology. The key challenges in this project are to (1) understand how system-level interactions (at both the hardware and software level) scale with system size, (2) devise scale models that are representative of the target system of interest, and (3) develop extrapolation models that are accurate, fast and easy to deploy. This project will propose and evaluate scale models for large multi-core CPUs (with multiple tens of cores) as well as high-end GPUs, and will consider a diverse range of workloads, including multi-program workloads, weak- and strong-scaling multi-threaded workloads and managed-language workloads. The evaluation will be done through simulation as well as on real hardware.