If you plan training a machine learning model on images or text data you will notice that state-of-the-art methods use deep learning algorithms like Convolutional Neural Networks or the Transformer. Those Deep learning algorithms, require extensive parallel processing capabilities that are best handled by Graphics Processing Units (GPUs). GPUs are designed to perform multiple computations simultaneously, making them significantly faster and more efficient for the parallel tasks that deep learning models demand.
For students without access to GPUs, there are several cloud-based platforms that offer free GPU resources. These services allow you to train complex models without the need for expensive hardware investments. On this page, we will introduce you to a selection of services that provide free access to GPUs, along with information on the number of GPU hours you can expect to receive.
GPU Models: NVIDIA Tesla P100 with 16 GB GPU memory or Dual Tesla T4 with 15 GB GPU memory.
Free GPU Access: At least 30 hours per week.
Key Features: Houses over 50,000 public datasets
GPU Models: Nvidia K80s or Tesla T4, if available, with up to 16 GB of memory.
Free GPU Access: Up to 12 hours per session.
Key Features: Easy to use, integrates with Google Drive, supports real-time collaboration.
Limitations: Only works with notebooks, shared GPU, 30 minutes of idle time stops runtime, not ideal for larger models due to memory issues, requires reauthentication for Google Drive access.
GPU Memory: 8 GB.
Free GPU Access: Maximum runtime of 6 hours per session, with long idle times between 1–6 hours.
Key Features: Offers 8 CPUs with 30GB RAM, public and shareable Jupyter notebooks.
Limitations: No private notebooks on free GPU, GPU subject to availability, low free storage space of 5GB.
If the computation resources provided by the free GPU services are not sufficient for your project, please reach out to Johannes. We can offer additional compute capacity on a private server to support your needs.