![]() It is recalled that the standard deviation is a measure of the amount of variation or dispersion of a set of values and, in the present case, these deviations correspond to the fact that not all the machine instructions of a processor consume the same energy, and to the out-of-order execution, cache hits, and other run-time optimizations. It is also possible to obtain the energy efficiency of the specific domain of the isolated processors (UC + ALU + FPU + GT + other circuits on the microprocessor chip), as detailed in Table 12 and represented in Figure 15, where bars are included to depict the variability of the results. One data source available from Intel Power Gadget is the power of the processor (CPU) (see Table 6), which can be combined with the performance ratios of execution ( R avr and R max) obtained with Linpack. As stated in, “transistors are still getting faster generation-to-generation but not at the same rate than what used to be achieved in the 90s, since the major emphasis in transistor design has now shifted from speed to limiting power consumption.” This forecast is no longer being fulfilled since transistors are reaching an atomic size for which the traditional physics models are no longer valid. ![]() Combining this property with Moore’s Law, it can be stated that over time the number of transistors in a dense integrated circuit (IC) doubles about every 24 months while the performance per watt does so approximately every 17 months. In this context, Dennard scaling holds, which states that if the density of the transistors is doubled, the power consumption stays approximately the same. Keep in mind that the power consumed depends on various factors, such as the density of the microprocessor’s transistors, whose growth has been determined by Moore’s Law. Note that the power consumed (W) by the processors does not depend directly on their performance rate, as might seem intuitive. These two tools are briefly described below. Additionally, with the use of the Power Gadget, other collateral measurements to the objective of this work are obtained, such as RDTSC (number of cycles), elapsed time, CPU frequency, CPU utilization (%), and package temperature, which are included because they are related to the energy consumption of the complete equipment and can be analyzed in more detail in future works. Furthermore, the simultaneous use of these two tools makes it possible to check the consistency between the measurements (waveforms, execution times, etc.) made with openZmeter and those made internally. ![]() The openZmeter is used to measure global energy consumption in the SUTs, from which the energy efficiency is calculated ( Section 3), while the Power Gadget allows to obtain separately the specific energy efficiency of the processors of the analyzed systems ( Section 4). Materials and Methodsīoth an external energy smart wattmeter (openZmeter) and a monitoring tool (Intel Power Gadget) based on the RAPL control register interface are used in this work. This paper ends with some conclusions in Section 5. On the other hand, Section 3 shows the experimental results obtained, which are interpreted and discussed in Section 4. This includes the description and justification of the reference application ( Section 2.1), the energy consumption measurement tools used ( Section 2.2), the platforms and processors under test ( Section 2.3), and the methodology used to perform the experiments ( Section 2.4). Now that the context of the work and related works has been defined, in Section 2 the materials and methods are presented. The actions and ideas that are being developed to achieve this objective have also been outlined, among which is the migration of scientific-technical applications to personal computers whenever the time requirements of the applications allow it. In accordance with the aforementioned objectives, this section of this paper ( Section 1) described the importance of reducing the energy consumption of computing systems and ICTs to achieve a sustainable world.
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |