MacBook Pro for Deep Learning? Let’s Try.

Original article can be found here (source): Artificial Intelligence on Medium

Deep Learning — Performance Comparison

Now I’ve gone ahead and ran models on both laptops to see how much a MacBook will fall behind. Before showing you the results, you might also be interested in the model architecture — it’s shown on the image below:

So nothing special here, I’ve used ReLU for an activation function for the hidden layers and logarithmic softmax for the output layer. Also, I’ve performed MaxPooling after both convolutional layers with a kernel size of 2×2 and stride od 2. And yes, almost forgot — the model was trained for 15 epochs.

It’s your average setup and will result in around 150,000 trainable parameters (weights and biases) that neural network needs to optimize for.

Now comes the part you’ve been waiting for — actual performance comparison! Here’s the chart:

Lenovo: 9 mins 36 seconds, MacBook: 10 mins 13 seconds

As you can see, the Lenovo laptop performed better, but not by much. I’ve expected more severe difference, primarily because Lenovo laptop has a “decent” GPU, and MacBook doesn’t.

Because the ventilation system is impeccable on the Lenovo laptop and mostly non-existent on the Mac, the Mac did get hot, so that’s another area you should consider.

As the only goal of this article is to research can MacBooks actually handle deep learning and how well they perform at it, here’s where I’ll stop.