Abstract:
In order to observe and understand the architectural behavior of applications and evaluate new techniques, computer architects often use simulation tools. Several cycle-a...Show MoreMetadata
Abstract:
In order to observe and understand the architectural behavior of applications and evaluate new techniques, computer architects often use simulation tools. Several cycle-accurate simulators have been proposed to simulate the operation of the processor on the micro-architectural level. However, an important step before adopting a simulator is its validation, in order to determine how accurate the simulator is compared to a real machine. This validation step is often neglected with the argument that only the industry possesses the implementation details of the architectural components. The lack of publicly available micro-benchmarks that are capable of providing insights about the processor implementation is another barrier. In this paper, we present the validation of a new cycle-accurate, trace-driven simulator, SiNUCA. To perform the validation, we introduce a new set of micro-benchmarks to evaluate the performance of architectural components. SiNUCA provides a controlled environment to simulate the micro-architecture inside the cores, the cache memory sub-system with multi-banked caches, a NoC interconnection and a detailed memory controller. Using our micro-benchmarks, we present a simulation validation comparing the performance of real Core 2 Duo and Sandy-Bridge processors, achieving an average performance error of less than 9%.
Date of Conference: 24-26 August 2015
Date Added to IEEE Xplore: 30 November 2015
ISBN Information: