Check if your performance intuition still works with CUDA