Which Amazon SageMaker algorithms can only use GPU for training?

Documentation for Amazon SageMaker built-in algorithms provides recommendations around choice of Amazon EC2 instances and whether given algorithm supports GPU or CPU devices.

Let's take Image Classification as an example. Here is a excerpt from online documentation:

For image classification, we support the following GPU instances for training: ml.p2.xlarge, ml.p2.8xlarge, ml.p2.16xlarge, ml.p3.2xlarge, ml.p3.8xlargeand ml.p3.16xlarge. We recommend using GPU instances with more memory for training with large batch sizes. However, both CPU (such as C4) and GPU (such as P2 and P3) instances can be used for the inference. You can also run the algorithm on multi-GPU and multi-machine settings for distributed training.

For more complex scenarios, such as Script or BYO Container modes, customers have flexibility to choose which device (GPU or CPU) to utilize for which operation. This is configured as part of their training scripts.