v26.04.00

What's Changed

🚨 Breaking Changes

• Update CCCL to 3.3.0 pre-release by @bdice in #981

🐛 Bug Fixes

• FINAL_CODEBLOCK can see targets that should be global by @robertmaynard in #993

🚀 New Features

• Add URL option to fetch packages from GitHub tarballs by @bdice in #968
• CMake 3.30+ means CMAKE_PROJECT_INCLUDE can be a list by @robertmaynard in #984
• Update cuco version to fetch the cuda fancy iterators by @PointKernel in #987

🛠️ Improvements

• Forward-merge release/26.02 into main by @bdice in #974
• Drop Python 3.10 support by @gforsyth in #975
• Use verify-hardcoded-version pre-commit hook by @KyleFromNVIDIA in #977
• Don't hard-code branch names by @KyleFromNVIDIA in #978
• Update CCCL to 3.2.1 pre-release by @bdice in #979
• Add support for Python 3.14 by @gforsyth in #982

Full Changelog: v26.04.00a...v26.04.00

v26.02.00

What's Changed

🚨 Breaking Changes

• Default to static linking of libcudart by @bdice in #949
• Update to CCCL v3.2.0 by @bdice in #955

🐛 Bug Fixes

• Update to a newer CCCL 3.2.x prerelease with a fix for environment queries by @bdice in #967

🚀 New Features

• Add support for finding CMake prefixes exported by Python packages by @vyasr in #951

🛠️ Improvements

• Use strict priority in CI conda tests by @bdice in #942
• Use strict priority in CI conda tests by @bdice in #947
• Upgrade to nvCOMP 5.1.0.21 by @bdice in #946
• Enable merge barriers by @KyleFromNVIDIA in #948
• feat(SABI): use stable abi naming when requested by @gforsyth in #954
• Bump cuco version to fetch HLL fix by @PointKernel in #952
• Clarify docs for rapids_cuda_init_architectures() by @KyleFromNVIDIA in #956
• Add CUDA 13.1 support by @bdice in #953
• Bump cuco version to fetch new HLL add_if API and bug fixes by @PointKernel in #958
• build and test against CUDA 13.1.0 by @jameslamb in #960
• Update to a newer CCCL 3.2.x prerelease with a fix for mdspan by @bdice in #959
• Update to a newer CCCL 3.2.x prelease by @nirandaperera in #961
• Use main shared-workflows branch by @jameslamb in #962
• Cherry-pick 25.12 changelog to main by @AyodeAwe in #963
• Fix update-version.sh to not modify cmake_minimum_required by @AyodeAwe in #964
• fix(build): build package on merge to release/* branch by @gforsyth in #971
• Update CCCL to final v3.2.0 release by @bdice in #970

Full Changelog: v26.02.00a...v26.02.00

v25.12.00

What's Changed

🚨 Breaking Changes

• Remove deprecated support for fmt and spdlog by @bdice in #914
• Remove deprecated support for CUDA 11 by @bdice in #913
• Remove spdlog and fmt and move to rapids-logger 0.2.1 by @vyasr in #918
• Require CUDA 12.2+ by @jakirkham in #934

🐛 Bug Fixes

• Fix RAPIDS_BRANCH file by @KyleFromNVIDIA in #910
• Temporarily restore spdlog and fmt by @vyasr in #917
• Update rapids-logger to 0.2.2 by @vyasr in #920
• Can't use Genex in GLOBAL_TARGETS by @robertmaynard in #921
• Update rapids-logger to 0.2.3 by @vyasr in #923
• refactored update-version.sh to handle new branching strategy by @rockhowse in #936

🚀 New Features

• Adding ENABLE_UNSTABLE option for rapids_cpm_cccl by @nirandaperera in #919

🛠️ Improvements

• Update nvbench by @bdice in #895
• Update to CCCL v3.1.0 by @bdice in #906
• Upgrade to CPM 0.42.0 by @vyasr in #916
• Forward-merge branch-25.10 into branch-25.12 by @bdice in #927
• Update CCCL to final v3.1.0 release tag by @bdice in #928
• Use main in RAPIDS_BRANCH by @bdice in #930
• Use main branch by @bdice in #931
• Use main shared-workflows branch by @bdice in #932
• Use SPDX for all copyright headers by @KyleFromNVIDIA in #933
• Bump cuco version to fetch stream-ordered allocator by @PointKernel in #929
• Bump cuco version to bypass equal check for multiset/multimap by @PointKernel in #935
• Some more SPDX fixes by @KyleFromNVIDIA in #937
• Update CCCL to get backports for XGBoost compatibility by @bdice in #944

New Contributors

• @nirandaperera made their first contribution in #919
• @rockhowse made their first contribution in #936

Full Changelog: v25.12.00a...v25.12.00

v25.10.00

🚨 Breaking Changes

• Upgrade to nvCOMP 5.0.0.6 (#896) @vuule

🐛 Bug Fixes

• Fix generation of RAPIDS_BRANCH (#909) @KyleFromNVIDIA
• change Architectures to avoid compute_100/120 code gen bug (#900) @robertmaynard
• set_architectures + NATIVE works even without a GPU (#899) @robertmaynard
• set_architectures correctly limits CUDA 12.8 to sm_120 (#898) @robertmaynard
• rapids_export handles PROJECT_VERSION CMake 4.1+ change in behavior (#893) @robertmaynard
• Fix detect architectures to support gcc 14 (#892) @ChuckHastings
• Update RAPIDS.cmake to match new logic in other projects (#884) @robertmaynard
• Update RAPIDS_BRANCH to 25.10 (#883) @robertmaynard
• Expand versions.json to have src_dir representation (#882) @robertmaynard

🚀 New Features

• Update cuco version to fetch signed comparison warning fix (#905) @PointKernel
• Updated dependencies.yaml to support CUDA 13 (#901) @robertmaynard
• Use family arch flags whenever possible (#897) @robertmaynard
• Upgrade to nvCOMP 5.0.0.6 (#896) @vuule
• Update set_architectures to support CUDA 13 (#891) @robertmaynard
• rapids-cmake Expose rapids_cuda_enable_fatbin_compression to have better CUDA 12/13 compression API (#890) @robertmaynard
• CUDA 13 builds will properly pull nvcomp 4.2.0.11 (#889) @robertmaynard
• Properly store source_subdir values from rapids_cpm_find calls (#885) @robertmaynard

🛠️ Improvements

• Configure repo for automatic release notes generation (#907) @AyodeAwe
• Use branch-25.10 again (#904) @jameslamb
• Update rapids-dependency-file-generator (#903) @KyleFromNVIDIA
• Update cuco version to fetch the new roaring bitmap (#902) @PointKernel
• Bump cuco version to fetch a bug fix for device retrieve APIs (#887) @PointKernel
• chore: update docker version tags (#873) @gforsyth

[NIGHTLY] v25.12.00

🔗 Links

• Development Branch
• Compare with main branch

🚨 Breaking Changes

• Remove spdlog and fmt and move to rapids-logger 0.2.1 (#918) @vyasr
• Remove deprecated support for fmt and spdlog (#914) @bdice
• Remove deprecated support for CUDA 11 (#913) @bdice

🐛 Bug Fixes

• Update rapids-logger to 0.2.3 (#923) @vyasr
• Can't use Genex in GLOBAL_TARGETS (#921) @robertmaynard
• Update rapids-logger to 0.2.2 (#920) @vyasr
• Temporarily restore spdlog and fmt (#917) @vyasr
• Fix RAPIDS_BRANCH file (#910) @KyleFromNVIDIA

🚀 New Features

• Adding ENABLE_UNSTABLE option for rapids_cpm_cccl (#919) @nirandaperera

🛠️ Improvements

• Update CCCL to final v3.1.0 release tag (#928) @bdice
• Forward-merge branch-25.10 into branch-25.12 (#927) @bdice
• Remove spdlog and fmt and move to rapids-logger 0.2.1 (#918) @vyasr
• Upgrade to CPM 0.42.0 (#916) @vyasr
• Remove deprecated support for fmt and spdlog (#914) @bdice
• Remove deprecated support for CUDA 11 (#913) @bdice
• Update to CCCL v3.1.0 (#906) @bdice
• Update nvbench (#895) @bdice

v25.08.00

🚨 Breaking Changes

• rapids_cpm_cccl: Remove support for CCCL < 2.8 (#859) @robertmaynard
• Remove CUDA 11 support (#855) @KyleFromNVIDIA
• Update to CCCL 3.0 (#854) @vyasr
• Require cpp subdirectory for RMM (#832) @bdice

🐛 Bug Fixes

• CCCL: disable PDL (#876) @bdice
• Use RMM main (new branching strategy) to fix downstream fetching issues (#862) @bdice
• rapids_cpm_cccl: Update to new location of cccl-config (#858) @robertmaynard
• Remove CCCL patches that aren't used anymore (#857) @robertmaynard
• Fetch the atomic fix in CCCL 3.0 (#856) @PointKernel

📖 Documentation

• add docs on CI workflow inputs (#868) @jameslamb

🚀 New Features

• Update CCCL version tag for PDL disable (#879) @davidwendt
• Use RAPIDS_BRANCH file to handle the new branching strategy (#870) @robertmaynard
• rapids-cmake: Add support for a version suffix to mean using main (#864) @robertmaynard
• rapids_cpm_cccl: Remove support for CCCL < 2.8 (#859) @robertmaynard
• Update to CCCL 3.0 (#831) @bdice

🛠️ Improvements

• Update to CCCL v3.0.2 (#878) @bdice
• fix(docker): use versioned -latest tag for all rapidsai images (#871) @gforsyth
• Revert "Use RMM main (new branching strategy)" (#869) @robertmaynard
• Rename *.hpp.in to *.h.in to signify that they are C headers (#867) @KyleFromNVIDIA
• refactor(shellcheck): enable for all files (#866) @gforsyth
• Remove nvidia and dask channels (#865) @vyasr
• Upgrade cuCollections to fetch the new storage for better runtime performance (#861) @PointKernel
• Remove CUDA 11 support (#855) @KyleFromNVIDIA
• Update to CCCL 3.0 (#854) @vyasr
• Deprecate fmt and spdlog (#853) @vyasr
• Forward-merge branch-25.06 into branch-25.08 (#848) @gforsyth
• Update to NVTX 3.2.0. (#844) @bdice
• Forward-merge branch-25.06 into branch-25.08 (#839) @gforsyth
• Temporarily use patched CCCL (#833) @bdice
• Require cpp subdirectory for RMM (#832) @bdice

v25.06.00

🚨 Breaking Changes

• Make RAPIDS.cmake inlineable without requiring updates (#825) @KyleFromNVIDIA
• Make RAPIDS.cmake inlineable without requiring updates (#822) @KyleFromNVIDIA
• Add build patch functionality (#801) @KyleFromNVIDIA

🐛 Bug Fixes

• Get newer CCCL 2.8 with fix for atomics. (#840) @bdice
• Unset CUDAARCHS in test (#835) @KyleFromNVIDIA
• Revert "Make RAPIDS.cmake inlineable without requiring updates" (#824) @bdice
• Add include_guard(GLOBAL) to download_with_retry.cmake (#818) @KyleFromNVIDIA
• Fix issue with how project pins are determined (#815) @KyleFromNVIDIA
• Disable git_shallow for nvtx3 and bs_thread_pool (#811) @KyleFromNVIDIA
• Fix NVTX3 export (#804) @KyleFromNVIDIA

🚀 New Features

• Upgrade nvbench for latest throttling control updates (#830) @PointKernel
• Add build patch functionality (#801) @KyleFromNVIDIA

🛠️ Improvements

• explicitly provide 'script' to CI workflows (#849) @jameslamb
• Finish CUDA 12.9 migration and use branch-25.06 workflows (#847) @bdice
• Update to clang 20 (#846) @bdice
• Update to CCCL 2.8.4. (#845) @bdice
• Bump cuco version to fetch multiple CCCL upgrade fixes (#843) @PointKernel
• Build and test with CUDA 12.9.0 (#842) @bdice
• Upgrade cuco version to include murmurhash performance fix (#838) @PointKernel
• feat: add Python 3.13 support (#834) @gforsyth
• Update cuCollections to avoid CCCL deprecation warnings (#828) @miscco
• Use cpp subdirectory for RMM. (#827) @bdice
• Make RAPIDS.cmake inlineable without requiring updates (#825) @KyleFromNVIDIA
• Remove hard-coded RAPIDS versions (#823) @KyleFromNVIDIA
• Make RAPIDS.cmake inlineable without requiring updates (#822) @KyleFromNVIDIA
• Format all CMake code (#821) @bdice
• In conda-build produced binaries, use paths relative to $PREFIX (#817) @jakirkham
• Remove CMP0135 policy checks. (#816) @bdice
• Use hashes instead of tags (#812) @KyleFromNVIDIA
• refactor(conda): remove unused conda package and associated CI resources (#810) @gforsyth
• Allow retrying downloads on network failures (#809) @pentschev
• Define -DCUB_DISABLE_NAMESPACE_MAGIC and -DCUB_IGNORE_NAMESPACE_MAGIC_ERROR in rapids_cpm_cccl (#807) @trxcllnt
• Update to CCCL 2.8.3 (#793) @bdice

[NIGHTLY] v25.08.00

🔗 Links

• Development Branch
• Compare with main branch

🚨 Breaking Changes

• rapids_cpm_cccl: Remove support for CCCL < 2.8 (#859) @robertmaynard
• Remove CUDA 11 support (#855) @KyleFromNVIDIA
• Update to CCCL 3.0 (#854) @vyasr
• Require cpp subdirectory for RMM (#832) @bdice

🐛 Bug Fixes

• Bump CCCL to v3.0.3 (#922) @Jacobfaib
• CCCL: disable PDL (#876) @bdice
• Use RMM main (new branching strategy) to fix downstream fetching issues (#862) @bdice
• rapids_cpm_cccl: Update to new location of cccl-config (#858) @robertmaynard
• Remove CCCL patches that aren't used anymore (#857) @robertmaynard
• Fetch the atomic fix in CCCL 3.0 (#856) @PointKernel

📖 Documentation

• add docs on CI workflow inputs (#868) @jameslamb

🚀 New Features

• Update CCCL version tag for PDL disable (#879) @davidwendt
• Use RAPIDS_BRANCH file to handle the new branching strategy (#870) @robertmaynard
• rapids-cmake: Add support for a version suffix to mean using main (#864) @robertmaynard
• rapids_cpm_cccl: Remove support for CCCL < 2.8 (#859) @robertmaynard
• Update to CCCL 3.0 (#831) @bdice

🛠️ Improvements

• Update to CCCL v3.0.2 (#878) @bdice
• fix(docker): use versioned -latest tag for all rapidsai images (#871) @gforsyth
• Revert "Use RMM main (new branching strategy)" (#869) @robertmaynard
• Rename *.hpp.in to *.h.in to signify that they are C headers (#867) @KyleFromNVIDIA
• refactor(shellcheck): enable for all files (#866) @gforsyth
• Remove nvidia and dask channels (#865) @vyasr
• Upgrade cuCollections to fetch the new storage for better runtime performance (#861) @PointKernel
• Remove CUDA 11 support (#855) @KyleFromNVIDIA
• Update to CCCL 3.0 (#854) @vyasr
• Deprecate fmt and spdlog (#853) @vyasr
• Forward-merge branch-25.06 into branch-25.08 (#848) @gforsyth
• Update to NVTX 3.2.0. (#844) @bdice
• Forward-merge branch-25.06 into branch-25.08 (#839) @gforsyth
• Temporarily use patched CCCL (#833) @bdice
• Require cpp subdirectory for RMM (#832) @bdice

v25.04.00

🚨 Breaking Changes

• Backport build patch functionality (#805) @KyleFromNVIDIA
• Update gtest version (#795) @vyasr
• Remove now superfluous fmt patch (#790) @vyasr
• Use CTest dynamic resource spec generation (#786) @KyleFromNVIDIA
• Switch to using rapids-logger as a library (#765) @vyasr
• Require CMake 3.30.4, and latest CPM (#753) @robertmaynard

🐛 Bug Fixes

• Set correct compiler feature for generate_ctest_json (#802) @KyleFromNVIDIA
• Also download fmt when building nvbench statically (#792) @vyasr
• Rollback to CPM 0.40.0 (#789) @robertmaynard
• Sanitize patch inputs so that quotes are properly escaped (#787) @robertmaynard
• Fix tests broken by rapids-logger changes (#775) @vyasr

🚀 New Features

• Backport build patch functionality (#805) @KyleFromNVIDIA
• Add support for components in Cython rapids_cython_create_modules (#783) @pentschev
• rapids_cpm_package_override: Always load overrides (#778) @robertmaynard
• Require CMake 3.30.4, and latest CPM (#753) @robertmaynard

🛠️ Improvements

• Update gtest version (#795) @vyasr
• Remove now superfluous fmt patch (#790) @vyasr
• Use CTest dynamic resource spec generation (#786) @KyleFromNVIDIA
• Use conda-build instead of conda-mambabuild (#784) @bdice
• Set -Wno-deprecated-gpu-targets when compiling for RAPIDS architectures (#776) @bdice
• Add shellcheck to pre-commit and fix warnings (#772) @gforsyth
• Update cuCollection to not rely on thrust::identity (#771) @miscco
• Use rapids-generate-version for package versions (#770) @bdice
• Use shared-workflows branch-25.04 (#768) @bdice
• add telemetry (#767) @msarahan
• Add build_type input field for test.yaml (#766) @gforsyth
• Switch to using rapids-logger as a library (#765) @vyasr
• Add verify-codeowners hook (#762) @KyleFromNVIDIA
• Forward-merge branch-25.02 to branch-25.04 (#760) @bdice
• Migrate to NVKS for amd64 CI runners (#758) @bdice

v25.02.01

🚨 Breaking Changes

• Update to CCCL 2.7.0-rc2. (#710) @bdice

🐛 Bug Fixes

• Sanitize patch input to protect against CMake Language limitations (#773) @robertmaynard
• Backport cccl PR 3578 to v2.7 to unblock CUDA 12.8 builds (#756) @robertmaynard
• Use sysroot 2.17 on CUDA < 11.8. (#745) @bdice
• rapids_cpm_cccl use proper CCCL version value to compute install rules (#742) @robertmaynard
• Correct nvbench + conda test to explicitly pull spdlog (#733) @robertmaynard

📖 Documentation

• fix signature in rapids_export() docs (#732) @jameslamb

🚀 New Features

• Update to nvcomp 4.2 which offers full blackwell support (#780) @robertmaynard
• Add support for new architectures in CUDA 12.8 (#746) @robertmaynard
• Update logger hash (#738) @vyasr
• Add rapids-logger (#737) @vyasr
• Fea/support embedded patch content (#728) @robertmaynard
• Update to CCCL 2.7.0-rc2. (#710) @bdice

🛠️ Improvements

• Revert CUDA 12.8 shared workflow branch changes (#759) @vyasr
• Build and test with CUDA 12.8.0 (#754) @bdice
• Allow sysroot 2.28 in CUDA 11 jobs. (#749) @bdice
• Fetch cuco filter alignment fix to unblock RAPIDS CI (#744) @PointKernel
• Use CCCL v2.7.0 final tag. (#743) @bdice
• Use GCC 13 in CUDA 12 conda builds. (#741) @bdice
• Add new global targets for rmm (#739) @vyasr
• remove setup.cfg, use pyproject.toml (#736) @jameslamb
• Bump cuco by one commit to include cuco::arrow_filter_policy updates (#735) @mhaseeb123
• Require Cython to be found (#734) @vyasr
• Update version references in workflow (#729) @AyodeAwe
• Bump cuco version to fetch various fixes and new features (#723) @PointKernel
• Add breaking change workflow trigger (#719) @AyodeAwe