Objective: Deployable Electrocardiogram (ECG) arrhythmia detection on standard Central Processing Units (CPUs) remains challenging due to deep learning models’ computational demands. This paper proposes a lightweight One-Dimensional Convolutional Neural Network (1D-CNN) optimized for CPU execution. Methods: The model (two convolutional layers, <10,000 parameters) evaluated on the MIT-BIH Arrhythmia Database using single-split (80/20) and 10-fold cross-validation protocols. Results: Single-split evaluation achieved 97.52% accuracy in 133 seconds. Cross-validation revealed 95.62% mean accuracy. (±8.51% SD) with 0.665 macro F1-score, processing each fold in 34.75 seconds (347.5 seconds total). The 300-second CPU constraint was satisfied for single-split but exceeded by 15.8% for cross-validation. Conclusion: The proposed model achieves an acceptable accuracy-efficiency trade-off for CPU-based screening applications. However, the 8.51% standard deviation across folds and persistent accuracy-F1 gap (95.62% vs. 0.665) indicate that patient-wise validation and class imbalance mitigation are necessary for diagnostic deployment. Significance: This work establishes a baseline for CPU-efficient ECG analysis while identifying critical methodological requirements.