Title: Extending Amdahl's Law for Energy-Efficient Computing in the Many-Core Era
Abstract: An updated take on Amdahl's analytical model uses modern design constraints to analyze many-core design alternatives.The revised models provide computer architects with a better understanding of manycore design types, enabling them to make more informed tradeoffs. Unsustainable power consumption and ever-increasing design and verification complexity have driven the microprocessor industry to integrate multiple cores on a single die, or multicore, as an architectural solution sustaining Moore's law. 1 With dual-core and quad-core processors on the market and oct-core on the horizon, researchers already are a step ahead.They're investigating architectures, compilers, and programming models for a many-core processor with hundreds or even 1,000 cores on a single platform. 2,3n 1967, Gene Amdahl proposed an often overlooked law of scaling: A program's sequential computation largely limits the maximum achievable speedup. 4This implies that any nonparallel execution or intercore communication will rapidly diminish the performance scalability for parallel applications regardless of the amount of additional computation resources.A simple, yet insightful, observation, Amdahl's law continues to serve as a guideline for parallel programmers to assess the upper bounds of attainable performance.Unfortunately, beyond performance, computer architects face another Grand Challenge: energy efficiency.Architects should carefully design a future many-core processor so that its power consumption doesn't exceed its power budget. 5For example, a 16core processor with each core consuming an average of 20 watts will lead to 320 watts total power when all cores are active.This level of consumption can easily exceed a single processor die's power budget.In other words, the amount of power each core consumes will dictate the number of cores architects can integrate on-die.Apparently, power is becoming more critical than performance in scaling up many-core processors.Thus, before integrating a large number of cores on-chip to provide desired performance and throughput, architects must maximize each core's power efficiency.Tackling these new design challenges requires extending Amdahl's law to account for power scalability's implications in the coming many-core era.As the original Amdahl's law demonstrates, a simple analytical model can provide computer architects with useful insights.By using simple analytical models at the early design phase, we aim to provide a better understanding of energy-efficiency's limits, some feasible many-core design options, and future directions for making many-core more scalable.