google-deepmind / google-deepmind/deepmind-research
TensorFlow 2.4.1 GPU is not compatible with the H100 GPU.
Open
- Dominant language
- Jupyter Notebook
- Stars
- 15.2k
- Forks
- 2.9k
- PR merge metrics
- No merged PRs in 30d
Description
Enformer requires TensorFlow 2.4.1, but this version is not compatible with the NVIDIA H100 GPU (Hopper architecture). The H100 GPU requires CUDA 11.8+ and cuDNN 8.6+, while TensorFlow 2.4.1 only supports CUDA 11.0 and cuDNN 8.0. This causes fallback to CPU execution or errors when running training scripts.
Question: Are there alternative package versions or modifications to the Enformer setup that would allow it to run effectively (training) on an H100 GPU?
Contributor guide
Assessment
This issue has not been assessed yet.