v25.08.00

🚨 Breaking Changes

• Update requirements to CUDA 12.0+ (#1984) @bdice
• Remove CUDA 11 from dependencies.yaml (#1934) @KyleFromNVIDIA
• stop uploading packages to downloads.rapids.ai (#1929) @jameslamb

🐛 Bug Fixes

• Temporarily disable failing test on HMM systems. (#1950) @bdice
• Fix race conditions and deadlocks in REPLAY_BENCH (#1940) @wence-

📖 Documentation

• Update Python build instructions to include librmm wheel (#1978) @gmarkall
• Fix Python path in CONTRIBUTING.md (#1936) @bdice

🚀 New Features

• Update requirements to CUDA 12.0+ (#1984) @bdice

🛠️ Improvements

• Use size_type in device_uvector (#1992) @bdice
• chore: remove unused line from update-version.sh (#1989) @gforsyth
• Revert "Update branches that trigger nightlies (#1954)" (#1988) @gforsyth
• fix(docker): use versioned -latest tag for all rapidsai images (#1987) @gforsyth
• Move more implementations to precompiled shared library (#1980) @bdice
• [pre-commit.ci] pre-commit autoupdate (#1979) @pre-commit-ci[bot]
• Add managed memory resource to replay benchmark (#1938) @pentschev
• Remove CUDA 11 from dependencies.yaml (#1934) @KyleFromNVIDIA
• Remove CUDA 11 devcontainers and update CI scripts (#1933) @bdice
• refactor(rattler): remove cuda11 options and general cleanup (#1932) @gforsyth
• stop uploading packages to downloads.rapids.ai (#1929) @jameslamb
• Forward-merge branch-25.06 into branch-25.08 (#1925) @gforsyth
• Branch 25.08 merge branch 25.06 (#1914) @vyasr
• Forward-merge branch-25.06 into branch-25.08 (#1905) @gforsyth

[NIGHTLY] v25.10.00

🔗 Links

• Development Branch
• Compare with main branch

🐛 Bug Fixes

• Fix aligned resource adaptor bug when alignment is less than 256. (#2003) @bdice
• Skip test on HMM systems with older CUDA drivers (#1944) @bdice

🚀 New Features

• Update to use CUDA runtime APIs that work in CUDA 12 and 13 (#2004) @robertmaynard

🛠️ Improvements

• Update rapids-build-backend to 0.4.1 (#2007) @KyleFromNVIDIA
• ci(labeler): update labeler action to @v5 (#2006) @gforsyth
• [StepSecurity] Apply security best practices (#2000) @stepsecurity-app[bot]
• Use std::for_each instead of std::equal in test (#1999) @vyasr
• Allow latest OS in devcontainers (#1997) @bdice
• Update build infra to support new branching strategy (#1994) @robertmaynard
• Move more implementations to precompiled shared library (part 2) (#1983) @bdice
• Use GCC 14 in conda builds. (#1963) @vyasr

v25.06.00

🚨 Breaking Changes

• Convert part of RMM to a precompiled library (#1896) @bdice
• Move RMM C++ code into cpp directory. (#1883) @bdice

🐛 Bug Fixes

• Run system MR tests in isolation. (#1945) @bdice
• Use auditwheel to properly retag the wheel (#1913) @vyasr
• Fix logger macros (#1884) @vyasr

📖 Documentation

• Move docs to top level. (#1917) @bdice
• Update Readme for the logging set_level (#1911) @JigaoLuo
• Fixed documentation example for DeviceBuffer.to_device (#1881) @TomAugspurger

🚀 New Features

• Convert part of RMM to a precompiled library (#1896) @bdice
• Set mempool hw_decompress flag if driver supports it (#1875) @bdice
• Expose option to enable fabric memory handle support to Python (#1787) @pentschev

🛠️ Improvements

• fix(pytest): disable warning that gets raised to INTERNALERROR in pytest8.4.0 (#1942) @gforsyth
• use 'rapids-init-pip' in wheel CI, other CI changes (#1926) @jameslamb
• Finish CUDA 12.9 migration and use branch-25.06 workflows (#1921) @bdice
• Update to clang 20 (#1918) @bdice
• Quote head_rev in conda recipes (#1915) @bdice
• Build and test with CUDA 12.9.0 (#1907) @bdice
• Fix cpp wheel name to librmm. (#1903) @bdice
• Revert "Publish wheels and conda packages from Github Artifacts" (#1898) @bdice
• Publish wheels and conda packages from Github Artifacts (#1897) @VenkateshJaya
• Download build artifacts from Github for CI (#1895) @VenkateshJaya
• remove mkdir and test corresponding shared workflow (#1892) @msarahan
• Revert "Auto-sync draft PRs" (#1891) @bdice
• Auto-sync draft PRs (#1890) @bdice
• Add ARM conda environments (#1889) @bdice
• Vendor RAPIDS.cmake to avoid network call. (#1886) @bdice
• [pre-commit.ci] pre-commit autoupdate (#1885) @pre-commit-ci[bot]
• Move RMM C++ code into cpp directory. (#1883) @bdice
• refactor(rattler): enable strict channel priority for builds (#1867) @gforsyth
• Add support for Python 3.13 (#1851) @bdice
• Streamlining wheel builds to use fixed location and uploading build artifacts to Github (#1810) @VenkateshJaya

v25.04.00

🚨 Breaking Changes

• Add OOM fail reason, attempted allocation size to exception messages (retry) (#1844) @pmattione-nvidia
• Use new rapids-logger library (#1808) @vyasr

🐛 Bug Fixes

• Revert "Set mempool hw_decompress flag if driver supports it (#1854)" (#1873) @wence-
• Fix run export on cudatoolkit (#1862) @vyasr
• Fix dependencies.yaml for update-version.sh (#1859) @raydouglass
• Embed __FILE__ as C-string for prefix replacement (#1858) @jakirkham
• Add OOM fail reason, attempted allocation size to exception messages (retry) (#1844) @pmattione-nvidia
• Revert "Add OOM fail reason, attempted allocation size to exception messages" (#1843) @pmattione-nvidia
• fix GITHUB_WORKSPACE not being present locally (#1841) @msarahan
• Add telemetry setup to build workflows (#1838) @bdice
• Use static gbench (#1837) @bdice
• Fixes for rattler recipe (#1835) @bdice
• Depend on rapids-logger in host to prevent redistribution (#1834) @bdice
• Add OOM fail reason, attempted allocation size to exception messages (#1827) @pmattione-nvidia

📖 Documentation

• mr/host: fix incorrect docs usage of device_memory_resource (#1809) @ghost

🚀 New Features

• Add async view memory resource bindings to Python. (#1864) @bdice
• Run examples in CI (#1850) @bdice
• Add tests for RMM internal macros. (#1847) @bdice
• Add basic example. (#1800) @bdice

🛠️ Improvements

• Set mempool hw_decompress flag if driver supports it (#1854) @wence-
• Error if LIBCUDACXX_ENABLE_EXPERIMENTAL_MEMORY_RESOURCE is not defined. (#1852) @bdice
• Fix for -fdebug-prefix-map breaking sccache (#1846) @bdice
• fix(rattler): force cuda_major and date_string to be strings (#1842) @gforsyth
• use gha-tools rapids-telemetry-setup for mkdir -p (#1839) @msarahan
• fix(rattler): resolve all overlinking errors (#1836) @gforsyth
• Update rattler-build recipe with assorted small fixes (#1832) @gforsyth
• Sccache stats telemetry (#1830) @msarahan
• Consolidate more Conda solves in CI (#1828) @KyleFromNVIDIA
• Require CMake 3.30.4 (#1826) @robertmaynard
• Create Conda CI test env in one step (#1824) @KyleFromNVIDIA
• Apply IWYU changes and fix deprecated GTest usage (#1821) @bdice
• Remove unnecessary index (#1820) @vyasr
• Use shared-workflows branch-25.04 (#1816) @bdice
• Use rapids-pip-retry in CI jobs that might need retries (#1814) @gforsyth
• Use nightly matrix for branch tests. (#1813) @bdice
• Use build_type input (#1812) @bdice
• Add build_type to workflow inputs (#1811) @gforsyth
• Use new rapids-logger library (#1808) @vyasr
• Forward-merge branch-25.02 to branch-25.04 (#1806) @bdice
• disallow fallback to Make in wheel builds (#1804) @jameslamb
• Migrate to NVKS for amd64 CI runners (#1803) @bdice
• Branch 25.04 merge branch 25.02 (#1799) @vyasr
• Port to rattler-build (#1796) @gforsyth

v25.02.00

🚨 Breaking Changes

• Switch to using separate rapids-logger repo (#1774) @vyasr
• Remove deprecated factory functions from resource adaptors. (#1767) @bdice
• Remove rmm._lib (#1765) @Matt711
• Remove memory access flags from cuda_async_memory_resource (#1754) @abellina
• Create logger wrapper around spdlog that can be easily reused in other libraries (#1722) @vyasr

🐛 Bug Fixes

• Add missing array header include (#1771) @robertmaynard
• Remove memory access flags from cuda_async_memory_resource (#1754) @abellina
• Update build.sh (#1749) @vyasr
• Fix some logger issues (#1739) @vyasr
• Use consistent signature for target_link_libraries (#1738) @vyasr

📖 Documentation

• Revise README. (#1747) @bdice

🚀 New Features

• Make the stream module a part of the public API (#1775) @Matt711
• Remove deprecated factory functions from resource adaptors. (#1767) @bdice
• Remove rmm._lib (#1765) @Matt711
• Reduce dependencies on numba. (#1761) @bdice
• Use ruff, remove isort and black. (#1759) @bdice
• Use bindings layout for all cuda-python imports. (#1756) @bdice
• Add configuration for pre-commit.ci, update pre-commit hooks (#1746) @bdice
• Adds fabric handle and memory protection flags to cuda_async_memory_resource (#1743) @abellina
• Remove upper bounds on cuda-python to allow 12.6.2 and 11.8.5 (#1729) @bdice

🛠️ Improvements

• Revert CUDA 12.8 shared workflow branch changes (#1805) @vyasr
• Build and test with CUDA 12.8.0 (#1797) @bdice
• Disable exec checks for device_uvector::operator= (#1790) @miscco
• Add upper bound to prevent usage of numba 0.61.0 (#1789) @galipremsagar
• Add shellcheck to pre-commit and fix warnings (#1788) @gforsyth
• Add spdlog back as a requirement for now (#1780) @vyasr
• [pre-commit.ci] pre-commit autoupdate (#1778) @pre-commit-ci[bot]
• Use rapids-cmake for the logger (#1776) @vyasr
• Switch to using separate rapids-logger repo (#1774) @vyasr
• Use GCC 13 in CUDA 12 conda builds. (#1773) @bdice
• Check if nightlies have succeeded recently enough (#1772) @vyasr
• Fix codespell behavior. (#1769) @bdice
• Remove ignored cuda-python deprecation warning. (#1768) @bdice
• Forward-merge branch-24.12 to branch-25.02 (#1766) @bdice
• Update version references in workflow (#1757) @AyodeAwe
• gate telemetry dispatch calls on TELEMETRY_ENABLED env var (#1752) @msarahan
• Update cuda-python lower bounds to 12.6.2 / 11.8.5 (#1751) @bdice
• remove certs and simplify telemetry summarize (#1750) @msarahan
• stop installing 'wheel' in wheel-building script (#1748) @jameslamb
• Require approval to run CI on draft PRs (#1737) @bdice
• Create logger wrapper around spdlog that can be easily reused in other libraries (#1722) @vyasr
• Add breaking change workflow trigger (#1719) @AyodeAwe

v24.12.00

🚨 Breaking Changes

• Deprecate support for directly accessing logger (#1690) @vyasr

🐛 Bug Fixes

• Query total memory in failure_callback_resource_adaptor tests (#1734) @harrism
• Treat deprecation warnings as errors and fix deprecation warnings in replay benchmark (#1728) @harrism
• Disallow cuda-python 12.6.1 and 11.8.4 (#1720) @bdice
• Fix typos in .gitignore (#1697) @charlesbluca
• Fix rmm ._lib imports (#1693) @Matt711

📖 Documentation

• Fix docs warning (#1706) @bdice
• Update cross-link to cuda-python object (#1699) @wence-

🚀 New Features

• Correct rmm tests for validity of device pointers (#1714) @robertmaynard
• Update rmm tests to use rapids_cmake_support_conda_env (#1707) @robertmaynard
• adding telemetry (#1692) @msarahan
• Make cudaMallocAsync logic non-optional as we require CUDA 11.2+ (#1667) @robertmaynard

🛠️ Improvements

• enforce wheel size limits, README formatting in CI (#1726) @jameslamb
• Remove all explicit usage of fmtlib (#1724) @harrism
• WIP: put a ceiling on cuda-python (#1723) @jameslamb
• use rapids-generate-pip-constraints to pin to oldest dependencies in CI (#1716) @jameslamb
• Deprecate rmm._lib (#1713) @Matt711
• print sccache stats in builds (#1712) @jameslamb
• [fea] Expose the arena mr to the Python interface. (#1711) @trivialfis
• devcontainer: replace VAULT_HOST with AWS_ROLE_ARN (#1708) @jjacobelli
• make conda installs in CI stricter (part 2) (#1703) @jameslamb
• Add BUILD_SHARED_LIBS option defaulting to ON (#1702) @wence-
• make conda installs in CI stricter (#1696) @jameslamb
• Prune workflows based on changed files (#1695) @KyleFromNVIDIA
• Deprecate support for directly accessing logger (#1690) @vyasr
• Use rmm::percent_of_free_device_memory in arena test (#1689) @wence-
• exclude 'gcovr' from list of development pip packages (#1688) @jameslamb
• [Improvement] Reorganize Cython to separate C++ bindings and make Cython classes public (#1676) @Matt711

v24.10.00

🚨 Breaking Changes

• Inline functions that return static references must have default visibility (#1653) @wence-
• Hide visibility of non-public symbols (#1644) @jameslamb
• Deprecate adaptor factories. (#1626) @bdice

🐛 Bug Fixes

• Add missing include to resource_ref.hpp (#1677) @miscco
• Remove the friend declaration with an attribute (#1669) @kingcrimsontianyu
• Fix build.sh clean to delete python build directory (#1658) @rongou
• Stream synchronize before deallocating SAM (#1655) @rongou
• Explicitly mark RMM headers with RMM_EXPORT (#1654) @robertmaynard
• Inline functions that return static references must have default visibility (#1653) @wence-
• Use tool.scikit-build.cmake.version (#1637) @KyleFromNVIDIA

📖 Documentation

• Recommend miniforge for conda install. (#1681) @bdice
• Fix docs cross reference in DeviceBuffer.prefetch (#1636) @bdice

🚀 New Features

• [FEA] Allow setting *_pool_size with human-readable string (#1670) @Matt711
• Update RMM adaptors, containers and tests to use get/set_current_device_resource_ref() (#1661) @harrism
• Deprecate adaptor factories. (#1626) @bdice
• Allow testing of earliest/latest dependencies (#1613) @seberg
• Add resource_ref versions of get/set_current_device_resource (#1598) @harrism

🛠️ Improvements

• Update update-version.sh to use packaging lib (#1685) @AyodeAwe
• Use CI workflow branch 'branch-24.10' again (#1683) @jameslamb
• Update fmt (to 11.0.2) and spdlog (to 1.14.1). (#1678) @jameslamb
• Attempt to address oom failures in test suite (#1672) @wence-
• Add support for Python 3.12 (#1666) @jameslamb
• Update rapidsai/pre-commit-hooks (#1663) @KyleFromNVIDIA
• Drop Python 3.9 support (#1659) @jameslamb
• Remove NumPy <2 pin (#1650) @seberg
• Hide visibility of non-public symbols (#1644) @jameslamb
• Update pre-commit hooks (#1643) @KyleFromNVIDIA
• Improve update-version.sh (#1640) @bdice
• Install headers into ${CMAKE_INSTALL_INCLUDEDIR} (#1633) @KyleFromNVIDIA
• Merge branch-24.08 into branch-24.10 (#1631) @jameslamb

v24.08.00

🚨 Breaking Changes

• Add a stack to the statistics resource (#1563) @madsbk

🐛 Bug Fixes

• Rename .devcontainers for CUDA 12.5 (#1615) @jakirkham
• Avoid accessing statistics_resource_adaptor stack top if it is empty (#1588) @harrism
• Avoid --find-links. (#1583) @bdice
• Fix test_python matrix (#1579) @KyleFromNVIDIA
• Allow anonymous user in devcontainer name (#1576) @bdice

📖 Documentation

• Instruct to create associated issue in PR template. (#1624) @harrism
• add rapids-build-backend to docs (#1614) @jameslamb
• Revert "Remove HTML builds of librmm (#1415)" (#1604) @bdice
• Add documentation for CPM usage (#1600) @pauleonix
• Update Thrust CMake Guide link in README.md (#1593) @pauleonix

🚀 New Features

• Prefetch resource adaptor (#1608) @bdice
• Add python wrapper for system memory resource (#1605) @rongou
• Refactor mr_ref_tests to not depend on MR base classes (#1589) @harrism
• Add system memory resource (#1581) @rongou
• Add rmm::prefetch() and DeviceBuffer.prefetch() (#1573) @harrism

🛠️ Improvements

• split up CUDA-suffixed dependencies in dependencies.yaml (#1627) @jameslamb
• Remove prefetch factory. (#1625) @bdice
• Use workflow branch 24.08 again (#1617) @KyleFromNVIDIA
• Build and test with CUDA 12.5.1 (#1607) @KyleFromNVIDIA
• skip CMake 3.30.0 (#1603) @jameslamb
• Add RMM_USE_NVTX cmake option to provide localized control of NVTX for RMM (#1602) @jlowe
• Use verify-alpha-spec hook (#1601) @KyleFromNVIDIA
• Avoid --find-links in wheel jobs (#1586) @jameslamb
• resolve dependency-file-generator warning, remove unnecessary rapids-build-backend configuration (#1582) @jameslamb
• Remove THRUST_WRAPPED_NAMESPACE and tests (#1578) @harrism
• Remove text builds of documentation (#1575) @vyasr
• ensure update-version.sh preserves alpha specs (#1572) @jameslamb
• Add available_device_memory to fetch free amount of memory on a GPU (#1567) @galipremsagar
• Add a stack to the statistics resource (#1563) @madsbk
• Use rapids-build-backend. (#1502) @bdice

v24.06.00

🚨 Breaking Changes

• Refactor polymorphic allocator to use device_async_resource_ref (#1555) @harrism
• Remove deprecated functionality (#1537) @harrism
• Remove deprecated cuda_async_memory_resource constructor that takes thrust::optional parameters (#1535) @harrism
• Remove deprecated supports_streams and get_mem_info methods. (#1519) @harrism

🐛 Bug Fixes

• rmm needs to link to nvtx3::nvtx3-cpp to support installed nvtx3 (#1569) @robertmaynard
• Make sure rmm wheel dependency on librmm is updated [skip ci] (#1565) @raydouglass
• Don't ignore GCC-specific warning under Clang (#1557) @aaronmondal
• Add publish jobs for C++ wheels (#1554) @vyasr
• Explicitly use the current device resource in DeviceBuffer (#1514) @wence-

📖 Documentation

• Allow specifying mr in DeviceBuffer construction, and document ownership requirements in Python/C++ interfacing (#1552) @wence-
• Fix Python install instruction (#1547) @wence-
• Update multi-gpu discussion for device_buffer and device_vector dtors (#1524) @wence-
• Fix ordering / heading levels in README.md and python example in guide.md (#1513) @harrism

🚀 New Features

• Add NVTX support and RMM_FUNC_RANGE() macro (#1558) @harrism
• Always use a static gtest (#1532) @robertmaynard
• Build C++ wheel (#1529) @vyasr
• Remove deprecated supports_streams and get_mem_info methods. (#1519) @harrism

🛠️ Improvements

• update copyright dates (#1564) @jameslamb
• Overhaul ops-codeowners (#1561) @raydouglass
• Adding support for cupy.cuda.stream.ExternalStream (#1559) @lilohuang
• Refactor polymorphic allocator to use device_async_resource_ref (#1555) @harrism
• add RAPIDS copyright pre-commit hook (#1553) @jameslamb
• Enable warnings as errors for Python tests (#1551) @mroeschke
• Remove header existence tests. (#1550) @bdice
• Only use functions in the limited API (#1545) @vyasr
• Migrate to {{ stdlib("c") }} (#1543) @hcho3
• Fix cuda11.8 nvcc dependency (#1542) @trxcllnt
• add --rm and --name to devcontainer run args (#1539) @trxcllnt
• Remove deprecated functionality (#1537) @harrism
• Remove deprecated cuda_async_memory_resource constructor that takes thrust::optional parameters (#1535) @harrism
• Make thrust_allocator deallocate safe in multi-device setting (#1533) @wence-
• Move rmm Python package to subdirectory (#1526) @vyasr
• Remove a file not being used (#1521) @galipremsagar
• Remove unneeded update-version.sh update (#1520) @AyodeAwe
• Enable all tests for arm arch (#1510) @galipremsagar

v24.04.00

🚨 Breaking Changes

• Accept stream argument in DeviceMemoryResource allocate/deallocate (#1494) @wence-
• Replace all internal usage of get_upstream with get_upstream_resource (#1491) @miscco
• Deprecate rmm::mr::device_memory_resource::supports_streams() (#1452) @harrism
• Remove deprecated rmm::detail::available_device_memory (#1438) @harrism
• Make device_memory_resource::supports_streams() not pure virtual. Remove derived implementations and calls in RMM (#1437) @harrism
• Deprecate rmm::mr::device_memory_resource::get_mem_info() and supports_get_mem_info(). (#1436) @harrism

🐛 Bug Fixes

• Fix search path for torch allocator in editable installs and ensure CUDA support is available (#1498) @vyasr
• Accept stream argument in DeviceMemoryResource allocate/deallocate (#1494) @wence-
• Run STATISTICS_TEST and TRACKING_TEST in serial to avoid OOM errors. (#1487) @bdice

📖 Documentation

• Pin to recent breathe, to prevent getting an unsupported sphinx version. (#1495) @bdice

🚀 New Features

• Replace all internal usage of get_upstream with get_upstream_resource (#1491) @miscco
• Add complete set of resource ref aliases (#1479) @nvdbaranec
• Automate include grouping using clang-format (#1463) @harrism
• Add get_upstream_resource to resource adaptors (#1456) @miscco
• Deprecate rmm::mr::device_memory_resource::supports_streams() (#1452) @harrism
• Remove duplicated memory_resource_tests (#1451) @miscco
• Change rmm::exec_policy to take async_resource_ref (#1449) @miscco
• Change device_scalar to take async_resource_ref (#1447) @miscco
• Add device_async_resource_ref convenience alias (#1441) @harrism
• Remove deprecated rmm::detail::available_device_memory (#1438) @harrism
• Make device_memory_resource::supports_streams() not pure virtual. Remove derived implementations and calls in RMM (#1437) @harrism
• Deprecate rmm::mr::device_memory_resource::get_mem_info() and supports_get_mem_info(). (#1436) @harrism
• Support CUDA 12.2 (#1419) @jameslamb

🛠️ Improvements

• Use conda env create --yes instead of --force (#1509) @bdice
• Add upper bound to prevent usage of NumPy 2 (#1501) @bdice
• Remove hard-coding of RAPIDS version where possible (#1496) @KyleFromNVIDIA
• Requre NumPy 1.23+ (#1488) @jakirkham
• Use rmm::device_async_resource_ref in multi_stream_allocation benchmark (#1482) @miscco
• Update devcontainers to CUDA Toolkit 12.2 (#1470) @trxcllnt
• Add support for Python 3.11 (#1469) @jameslamb
• target branch-24.04 for GitHub Actions workflows (#1468) @jameslamb
• [FEA]: Use std::optional instead of thrust::optional (#1464) @miscco
• Add environment-agnostic scripts for running ctests and pytests (#1462) @trxcllnt
• Ensure that ctest is called with --no-tests=error. (#1460) @bdice
• Update ops-bot.yaml (#1458) @AyodeAwe
• Adopt the rmm::device_async_resource_ref alias (#1454) @miscco
• Refactor error.hpp out of detail (#1439) @lamarrr