> ## Documentation Index
> Fetch the complete documentation index at: https://www.thundercompute.com/docs/llms.txt
> Use this file to discover all available pages before exploring further.

# Compatibility

> Review known Thunder Compute compatibility limitations, including managed memory, CUDA MPS, FFmpeg, and GPU profilers.

Thunder Compute supports common CUDA and GPU development workflows, but some workloads and tools are not currently compatible with the platform.

## Known Incompatibilities

* **Managed memory / UVM**: try a library or implementation that does not rely on managed memory, such as switching from TensorFlow to PyTorch.
* **CUDA Multi-Process Service (MPS)**: run workloads without MPS, or use a standard multi-GPU instance when you need process-level GPU sharing.
* **FFmpeg GPU acceleration**: run FFmpeg without GPU acceleration.
* **GPU profilers**: use application-level logging, framework metrics, or benchmark scripts instead.

## Support

If you run into another compatibility issue, please [contact support](https://www.thundercompute.com/contact) with the command you ran, the error output, and any relevant logs.