The Midlifekicker Microarchitecture Evaluation Metric Stamatis Vassiliadis Computer Engineering TU Delft the Netherlands stamatis@dutepp0.et.TUDelft.nl http://ce.et.tudelft.nl Leonel Sousa Electrical and Computer Engineering IST/INESC-ID Lisboa, Portugal las@inesc-id.pt http://sips.inesc-id.pt Georgi N. Gaydadjiev Computer Engineering TU Delft the Netherlands georgi@dutepp0.et.TUDelft.nl http://ce.et.tudelft.nl Abstract We introduce the midlifekicker metric for evaluating mi- croarchitectures mostly during the design process. We assume a microarchitecture designed at a time T-1 and estimate if a new microarchitecture projected for time T has advantages over the microarchitecture designed at T-1 and remapped on the same technology at time T. We con- sider that microarchitects minimize the product cycles per instruction (CPI) x cycle time and estimate performance based on CPI with a ”soft-threshold” to include cycle time product effects. Some measurements are also reported. Keywords: ILP, microarchitecture, pipeline. 1 Introduction For a given architecture 1 performance improvements of processor products are derived mainly from technology, mi- croarchitectures, and logic design. In this paper we consider primary microarchitectures (also called a ”design point”) and propose a metric as a means to evaluate them. It is well understood that architectures, technologies and logic design techniques change (advance) on time making mi- croarchitectures to have an ”expiration” date. The time pe- riod spawned between the product release based on a spe- cific microarchitecture and the product released of the same architecture based on a completely new microarchitecture it is called the life of a microarchitecture. In this paper, based 1 We assume the original definition described for example in [3] here and in the rest of the presentation. on real world design experience, we address microarchitec- ture related questions including the following: • Is the microarchitecture to be developed better than the previous design point? • Can we determine if the performance due to the mi- croarchitecture alone is improving through the years? The approach we propose was used by one of the au- thor’s team to estimate microarchitecture performance and it is termed the midlifekicker metric. The name, midlife- kicker, is the name given by designers to a microarchitec- ture remapped to a technology other than the one originally planned and before its life expiration. This paper is organized as follows. Section 2 introduces the metric and the justification for the approach. In section 3 the considered processors and the experimental setup are presented. The CPI results experimentally obtained with the SPEC2000 benchmark programs for the various processors are also reported. Finally, section 4 concludes the paper. 2 The Midlifekicker Metric To take advantage of rapid technology improvements it is rather common in industrial practice to remap a design before a new design is made publicly available as a product. In the context of our discussion this means that a design is moved to a different (newer) technology (platform) than the one it was originally developed for. This common practice is possible because of the following conjunctures: • It is well known that the number of gates per chip is increasing, consequently the design will fit to at least