Modeling and prediction of thermal dissipation in heterogeneous CPU platforms
Öhrling, Joel (2020)
Öhrling, Joel
Åbo Akademi
2020
Julkaisu on tekijänoikeussäännösten alainen. Teosta voi lukea ja tulostaa henkilökohtaista käyttöä varten. Käyttö kaupallisiin tarkoituksiin on kielletty.
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi-fe2020051229533
https://urn.fi/URN:NBN:fi-fe2020051229533
Tiivistelmä
For 5G, the next generation of telecommunications technology, energy efficiency is an important aspect. It has been shown that cooling and heat management accounts for a large part of the energy consumed by the telecommunications infrastructure. Therefore, there is a need to address how heat dissipation can be limited and the energy spent on cooling reduced. The goal of this thesis is to explore thermal modeling approaches for multi-core heterogeneous processors and investigate which approaches can be utilized to predict the thermal dissipation with the highest accuracy. Previously proposed approaches to thermal modeling and system identification have been investigated and reviewed. Based on this, three approaches have been selected for implementation and compared in terms of prediction accuracy: a linear state-space identification approach using polynomial regressors, a NARX neural network approach and a recurrent neural network approach configured in an FIR model structure. These modeling approaches were each assessed for both 1 and 6 hours of training data collected from a multi-core heterogeneous ARM processor. The results showed that the state-space model based on polynomial regressors significantly outperformed the other two modeling approaches when trained with 1 hour of data. When the models were trained with 6 hours of data, all three modeling approaches yielded good results. However, the state-space approach still produced the lowest prediction error of the approaches.