# major release. Example: sm_80 kernels can run on sm_89 GPUs but
# not on sm_90 GPUs. compute_80 kernels though can also run on sm_90 GPUs.
build:cuda_clang --repo_env=TF_CUDA_COMPUTE_CAPABILITIES="sm_60,sm_70,sm_80,sm_89,compute_90"

# Set up compilation CUDA version and paths and use the CUDA Clang toolchain.
build:cuda_clang_official --config=cuda_clang
build:cuda_clang_official --action_env=TF_CUDA_VERSION="12"

Contributor explicitly and finally terminates Your grants, and (b) on an
ongoing basis, if such Contributor fails to notify You of the
non-compliance by some reasonable means prior to 60 days after You have
come back into compliance. Moreover, Your grants from a particular
Contributor are reinstated on an ongoing basis if such Contributor
notifies You of the non-compliance by some reasonable means, this is the

    is no longer experimental and allows the use of 16-bit floating point
    formats during training, improving performance by up to 3x on GPUs and 60%
    on TPUs. Please see below for additional details.

*   TensorFlow Profiler now supports profiling `MultiWorkerMirroredStrategy` and
    tracing multiple workers using the