Skip to content
This repository was archived by the owner on May 28, 2024. It is now read-only.
This repository was archived by the owner on May 28, 2024. It is now read-only.

TF 2.5+ GPU build no longer works because CUDA 11 configure is not supported #105

@trevor-m

Description

@trevor-m

With TensorFlow 2.5, there have been some code changes which require CUDA 11. You can read more about the issues when trying to build TF 2.5 against CUDA 10 here:
tensorflow/tensorflow#49983
tensorflow/tensorflow#48468

The build rules to configure CUDA in this custom-op repo (https://github.com/tensorflow/custom-op/tree/master/gpu) have not been updated in a long time. Because of that, it is not possible to compile against newer CUDA versions so we can no longer build our custom gpu plugin against newer TF releases.

Questions

  • Will this repository be updated to support newer versions of CUDA? Otherwise, it is impossible to build custom gpu ops with newer TF versions (2.5+).
  • There is an alternative way to build custom ops through using the main TensorFlow repository by placing the custom op code in this folder https://github.com/tensorflow/tensorflow/tree/master/tensorflow/core/user_ops which will reuse the configuration code from the main tensorflow repository. Is this the recommended approach moving forward and will this repository be deprecated? Is this approach equivalent in terms of what kind of plugins can be built?

Thanks!

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions