Predictive Modeling for CPU, GPU, and FPGA Performance and Power Consumption: A Survey

CPUs and dedicated accelerators (namely GPUs and FPGAs) continue to grow increasingly large and complex to support todays demanding performance and power requirements. Designers are tasked with evaluating the performance and power of similarly increasingly large design spaces during pre-silicon design for CPUs and GPUs to reduce time-to-market and limit manufacturing costs, or to figure out how to best map applications onto FPGAs using high-level synthesis tools. Typically, cycle-accurate simulators are used to evaluate workloads for pre-silicon CPUs and GPUs and to avoid the overhead of synthesis and place-and-route when targeting FPGAs; however, simulators exhibit prohibitively long run times that limit the number of design points and workloads that can be evaluated in a reasonable timeframe. This survey focuses on predictive modeling as an alternative to cycle-accurate simulation, which enables rapid evaluation of workloads and design points. When applied properly, predictive modeling can improve time to market, and can facilitate more comprehensive design space explorations with far less overhead than simulation. The survey focuses on predictive models applied to CPUs, GPUs, and FPGAs, noting that the general approach has been applied to many other computing platforms as well.