Single GPU training

GPUs dramatically increase training speed compared to CPUs by parallelizing computations in deep learning models.

To learn more about GPUs, read this guide.

Switch from CPU to GPU

Start by debugging and iterating your model on a CPU. Once it’s working, shift to a GPU to speed up the process without extra costs. In Studios, switching is easy and cost-effective.

For seamless transitions across CPUs and GPUs, consider utilizing our PyTorch Lightning library. It supports fine-tuning and pre-training on any hardware with zero code changes.

Monitor GPU use

The top of the Studio shows metrics about the GPU. If the GPU RAM isn’t fully used or the utilization isn’t constantly at 100%, adjust your model's hyperparameters to utilize the full GPU capacity.