Choosing the right type of hardware for deep learning tasks is a widely discussed topic. An obvious conclusion is that the decision should be dependent on the task at hand and based on factors such as throughput requirements and cost.