Intel's x86 architecture has been exposed to high resolution and stealthy cache side channel attacks (CSCAs) over past few years. In this paper, we present a novel technique to detect CSCAs on Intel's x86 architecture. The proposed technique comprises of multiple machine learning models that use real-time behavioral data of concurrent processes collected through Hardware Performance Counters (HPCs). In this work, we demonstrate that machine learning models, when coupled with intelligent performance monitoring of concurrent processes at hardware-level, can be used in security for early-stage detection of high precision and stealthier CSCAs. We provide extensive experiments with four variants of the state-of-the-art CSCAs. We demonstrate that our proposed technique is resilient to noise generated by the system under various loads. To do so, we provide results under realistic system load conditions with an evaluation metric comprising of detection accuracy, speed, system-wide performance overhead and confusion matrix for machine learning models. In experiments, our technique achieves detection accuracy of up to 99.51% for Flush+Reload attack on RSA, incurring a performance overhead of 1.63% and 99.99% accuracy on AES while incurring a maximum performance overhead of 8.28%. Our experimental results show consistency for Flush+Flush attack on different implementations of AES as well.