diff --git a/recipe/conda_build_config.yaml b/recipe/conda_build_config.yaml index 0ded3457e0..6a924c2d00 100644 --- a/recipe/conda_build_config.yaml +++ b/recipe/conda_build_config.yaml @@ -310,7 +310,7 @@ console_bridge: # temporarily zipped with cuda_compiler_version cudnn: - None - - '9' # [((linux and not ppc64le) or win64) and os.environ.get("CF_CUDA_ENABLED", "False") == "True"] + - 9.10 # [((linux and not ppc64le) or win64) and os.environ.get("CF_CUDA_ENABLED", "False") == "True"] cutensor: - 2 curl: diff --git a/recipe/migrations/cudnn910.yaml b/recipe/migrations/cudnn910.yaml deleted file mode 100644 index 0a657ef152..0000000000 --- a/recipe/migrations/cudnn910.yaml +++ /dev/null @@ -1,34 +0,0 @@ -# needs to be ordered before CUDA 13.0 migrator -migrator_ts: 1755016000 -__migrator: - kind: - version - migration_number: - 1 - build_number: - 1 - ordering: - cudnn: - - None - - 9 - - 9.10 - commit_message: | - Rebuild for cudnn 9.10 - - Conda-forge assumed cudnn 9.x builds would stay compatible with each other. - This is the case for the API/ABI of the library, but nvidia dropped support - for older GPU architectures in cudnn 9.11. Since we have no package-level - metadata about compatibility with specific GPU architectures, this effectively - breaks all packages built atop cudnn for users on affected older GPUs. - - In order to remedy this situation, we need to rebuild all cudnn-dependent - feedstocks against cudnn 9.10 (the last version with full architecture support), - before we mark all those newer cudnn builds as broken. This only affects artefacts - for CUDA 12.x; those for CUDA 13.x are not affected, since CUDA 13 never supported - those older architectures in the first place. - - For more details see: https://github.com/conda-forge/cudnn-feedstock/issues/124 - -cudnn: - - None - - 9.10 # [((linux and not ppc64le) or win64) and os.environ.get("CF_CUDA_ENABLED", "False") == "True"]