Owing to its associated recurrent seizures, epilepsy, a prevalent neurological disorder, significantly impairs the patient's quality of life. Effective automated seizure prediction can revolutionize epilepsy management and improve patient wellbeing. This study introduces a novel, non-patient-specific seizure prediction system powered by Graded Spiking Neural Networks (GSNNs) and optimized for Intel's Loihi 2 neuromorphic processor. Training this innovative predictor with the widely used CHB-MIT electroencephalogram (EEG) dataset yielded promising results, outperforming traditional EEG seizure-predictio
n methodologies. Further enhancements in the prediction performance were achieved through 1) hyperparameter optimization, 2) EEG channel selection to reduce the data volume, and 3) the introduction of a time-windowed voting (TWV) mechanism for increased robustness against noise and artifacts. The system was then implemented on Intel's Loihi 2 neuromorphic processor, and its performance was analyzed. When contrasted with traditional Artificial Neural Networks (ANNs) based systems, the GSNN-based predictor demonstrated superior efficiency and reduced computational complexity, underscoring the substantial potential of GSNNs in this domain.