Nowadays, computer networks serve as the foundation for many systems. As a result, they become exposed to attackers who may use them to obtain unauthorized access. And unfortunately, the variety of attacks makes classic systems unable to deal with all malicious traffics. Swarm Intelligence algorithms prove their effectiveness in solving several problems, even when time is a constrained, it can be adapted to GPUs to improve their efficiency. This paper proposes an interesting application of a novel swarm intelligence algorithm known as the Artificial Orca algorithm (AOA) to deal with detecting intrusions incoming in a system. The problem is called Intrusion Detection System, and the AOA is proposed as an optimization algorithm for Support Vector Machine (SVM-AOA) to classify traffic. Unlike the existing literature, this paper integrates GPU architecture to parallelize the AOA algorithm (GPU-SVM-AOA). To come out with the best/near-best parameters of the SVM model, the proposed methods have been subject to the NSL-KDD dataset, and several experiments were conducted under C++ and Cuda programming languages using the OpenCV library for the machine learning model. The outcomes show that the RBF function is the best kernel function for the NSL-KDD dataset. Comparing SVM-AOA and GPU-SVM-AOA with state-of-the-art methods, namely SVM-Particle Swarm Optimization, K-Nearest-Neighbor, and decision tree, show that the present contributions outperform especially GPU-SVM-AOA. In addition to being comparable to SVM-AOA in terms of accuracy, recall and fitness value, GPU-SVM-AOA also improves the running time.