The study investigates the impact of computational hardware on the training time of neural networks, comparing GPU and CPU performance in a binary classification problem. Performed experiments utilizing CUDA technology for GPU acceleration. The hardware specifications are thoroughly examined, covering local GPU and CPU configurations and cloud-based solutions, including Google Colab and Microsoft Azure. Experimental results compare the training times across local and cloud hardware. The findings demonstrate the significant advantages of GPUs in accelerating training, particularly for complex models and large datasets.