Compatibility
Below are some notes on Compatibility of different libraries with Thunder Compute
Use cases
Thunder Compute is optimized for AI development, production inference, and data science. Thunder Compute is the best option for bursty GPU workloads because you never pay for idle GPU time. This is especially beneficial in prototyping workflows with debugging time, or for inference servers that frequently sit idle between requests. That said, Thunder Compute has the full functionality of an EC2 instance and can be used as a complete replacement.
Officially supported libraries
The following libraries and tools are officially supported:
- PyTorch
- TensorFlow
- JAX
- PyTorch Lightning
- Hugging Face
- CuPy
Note: make sure you install the cuda-compatible version of these libraries, just like you would with a physical GPU e.g., tensorflow[and-cuda]. The cuda-compatible PyTorch binary is pre-installed on every Thunder Compute instance.
You can create and manage Thunder Compute instances with the tnr command line interface from any major operating system. Within a Thunder Compute instance, The following interfaces are officially supported:
- Jupyter Notebooks
- IPython
- Anaconda
- Docker
Known incompatibilities
Currently, Thunder Compute lacks official support for the following use cases. If you would like to run one of the following on Thunder Compute, please contact us and we will try our best to help!
- Graphics workloads (OpenGL, Vulkan, Blender, others)
- Multi-node training
Cryptocurrency mining
Mining, staking, or otherwise interacting with cryptocurrency is strictly prohibited on Thunder Compute. If cryptocurrency-related activity is detected, the associated account is immediately banned from Thunder Compute and the monthly billing credit is revoked. The account is then billed for the full amount of usage.