NVIDIA · bawr · Jun 11, 2026 · Jun 16, 2026 · Jun 16, 2026
diff --git a/.github/workflows/build-release.yml b/.github/workflows/build-release.yml
@@ -0,0 +1,205 @@
+name: Build / Release
+
+on:
+  workflow_dispatch:
+    inputs:
+      architecture:
+        description: Package architecture to build
+        required: true
+        default: amd64
+        type: choice
+        options:
+          - amd64
+          - aarch64
+      package_release:
+        description: Package release suffix
+        required: true
+        default: "2"
+        type: string
+      run_tests:
+        description: Run build-time tests
+        required: true
+        default: false
+        type: boolean
+      release_comments:
+        description: Optional release comments. Defaults to the build commit hash.
+        required: false
+        type: string
+
+permissions:
+  contents: write
+
+concurrency:
+  group: build-release-${{ github.ref }}
+  cancel-in-progress: false
+
+jobs:
+  packages:
+    name: Build packages and create release
+    runs-on: ubuntu-24.04
+    env:
+      DCGM_PACKAGE_RELEASE: ${{ inputs.package_release }}
+
+    steps:
+      - name: Check out repository
+        uses: actions/checkout@v4
+        with:
+          lfs: false
+
+      - name: Prepare release metadata
+        id: metadata
+        shell: bash
+        env:
+          PACKAGE_ARCHITECTURE: ${{ inputs.architecture }}
+          RELEASE_COMMENTS: ${{ inputs.release_comments }}
+        run: |
+          set -euo pipefail
+
+          version="$(sed -nE 's/^project\(datacenter-gpu-manager-4 VERSION ([^)]+)\)/\1/p' CMakeLists.txt)"
+          if [[ -z "$version" ]]; then
+            echo "Unable to determine DCGM version from CMakeLists.txt" >&2
+            exit 1
+          fi
+
+          tag="v${version}-${DCGM_PACKAGE_RELEASE}-${PACKAGE_ARCHITECTURE}"
+          name="DCGM ${version}-${DCGM_PACKAGE_RELEASE}-${PACKAGE_ARCHITECTURE}"
+
+          if [[ -n "$RELEASE_COMMENTS" ]]; then
+            body="$RELEASE_COMMENTS"
+          else
+            body="Build commit: ${GITHUB_SHA}"
+          fi
+
+          {
+            echo "version=$version"
+            echo "tag=$tag"
+            echo "name=$name"
+            echo "body<<EOF"
+            echo "$body"
+            echo "EOF"
+          } >> "$GITHUB_OUTPUT"
+
+      - name: Prepare Git LFS cache key
+        shell: bash
+        run: |
+          set -euo pipefail
+          git lfs install --local
+          git lfs ls-files -l | sort > .git-lfs-cache-key
+
+      - name: Cache Git LFS objects
+        uses: actions/cache@v4
+        with:
+          path: .git/lfs
+          key: git-lfs-${{ runner.os }}-${{ hashFiles('.git-lfs-cache-key') }}
+          restore-keys: |
+            git-lfs-${{ runner.os }}-
+
+      - name: Pull Git LFS objects
+        shell: bash
+        run: git lfs pull
+
+      - name: Set build architecture
+        id: architecture
+        shell: bash
+        env:
+          PACKAGE_ARCHITECTURE: ${{ inputs.architecture }}
+        run: |
+          set -euo pipefail
+
+          case "$PACKAGE_ARCHITECTURE" in
+            amd64)
+              docker_architecture=x86_64
+              package_architecture=amd64
+              ;;
+            aarch64)
+              docker_architecture=aarch64
+              package_architecture=aarch64
+              ;;
+            *)
+              echo "Unsupported architecture: $PACKAGE_ARCHITECTURE" >&2
+              exit 1
+              ;;
+          esac
+
+          {
+            echo "docker_architecture=$docker_architecture"
+            echo "package_architecture=$package_architecture"
+          } >> "$GITHUB_OUTPUT"
+
+      - name: Set up Docker Buildx
+        uses: docker/setup-buildx-action@v3
+
+      - name: Build cached DCGM build image
+        working-directory: dcgmbuild
+        shell: bash
+        run: |
+          set -euo pipefail
+
+          docker_architecture="${{ steps.architecture.outputs.docker_architecture }}"
+
+          docker buildx bake "dcgmbuild-${docker_architecture}" \
+            --set "common-host-software.cache-from=type=gha,scope=dcgmbuild-common-host-software" \
+            --set "common-host-software.cache-to=type=gha,scope=dcgmbuild-common-host-software,mode=max" \
+            --set "toolchain-${docker_architecture}.cache-from=type=gha,scope=dcgmbuild-toolchain-${docker_architecture}" \
+            --set "toolchain-${docker_architecture}.cache-to=type=gha,scope=dcgmbuild-toolchain-${docker_architecture},mode=max" \
+            --set "dcgmbuild-${docker_architecture}.cache-from=type=gha,scope=dcgmbuild-${docker_architecture}" \
+            --set "dcgmbuild-${docker_architecture}.cache-to=type=gha,scope=dcgmbuild-${docker_architecture},mode=max" \
+            --set "dcgmbuild-${docker_architecture}.output=type=docker"
+
+      - name: Build packages
+        env:
+          DCGM_SKIP_LFS_INSTALL: 1
+          DCGM_SKIP_PYTHON_LINTING: 1
+          NPROC: 4
+          RUN_TESTS: ${{ inputs.run_tests }}
+        shell: bash
+        run: |
+          set -euo pipefail
+
+          test_arguments=(--no-tests)
+          if [[ "$RUN_TESTS" == "true" ]]; then
+            test_arguments=()
+          fi
+
+          ./build.sh \
+            --release \
+            --arch "${{ steps.architecture.outputs.package_architecture }}" \
+            "${test_arguments[@]}" \
+            --deb \
+            --rpm \
+            --packages \
+            --package-release "$DCGM_PACKAGE_RELEASE"
+
+      - name: Collect packages
+        id: packages
+        shell: bash
+        run: |
+          set -euo pipefail
+
+          mkdir -p release-assets
+          find _out \
+            \( -name '*.deb' -o -name '*.ddeb' -o -name '*.rpm' -o -name '*.tar.gz' \) \
+            -type f \
+            -exec cp --target-directory release-assets {} +
+
+          if ! compgen -G 'release-assets/*' > /dev/null; then
+            echo "No package artifacts were produced" >&2
+            exit 1
+          fi
+
+      - name: Upload package artifacts
+        uses: actions/upload-artifact@v4
+        with:
+          name: dcgm-packages-${{ steps.architecture.outputs.package_architecture }}
+          path: release-assets/*
+          if-no-files-found: error
+
+      - name: Create GitHub release
+        uses: softprops/action-gh-release@v2
+        with:
+          tag_name: ${{ steps.metadata.outputs.tag }}
+          name: ${{ steps.metadata.outputs.name }}
+          body: ${{ steps.metadata.outputs.body }}
+          target_commitish: ${{ github.sha }}
+          fail_on_unmatched_files: true
+          files: release-assets/*
diff --git a/build.sh b/build.sh
@@ -44,6 +44,7 @@ Usage: ${0} [options] [-- [any additional cmake arguments]]
         -p --packages                                      : Generate tar.gz packages once the build is done
            --deb                                           : Generate *.deb packages once the build is done
            --rpm                                           : Generate *.rpm packages once the build is done
+           --package-release <release>                     : Override the package release suffix for DEB/RPM packages
         -c --clean                                         : Make clean rebuild
         -a --arch <arch>                                   : Make build for specified architecture. Supported are: amd64, aarch64
         -n --no-tests                                      : Do not run build-time tests
@@ -78,7 +79,7 @@ Usage: ${0} [options] [-- [any additional cmake arguments]]
           sanitizers compatibility."
 }
 
-LONGOPTS=address-san,arch:,clean,coverage,deb,debug,debug-find,debug-find-pkg:,gcc-analyzer,help,leak-san,no-install,no-tests,packages,release,rpm,thread-san,ub-san,vmware
+LONGOPTS=address-san,arch:,clean,coverage,deb,debug,debug-find,debug-find-pkg:,gcc-analyzer,help,leak-san,no-install,no-tests,package-release:,packages,release,rpm,thread-san,ub-san,vmware
 SHORTOPTS=drsa:pchn
 
 ! PARSED=$(getopt --options=${SHORTOPTS} --longoptions=${LONGOPTS} --name "${0}" -- "$@")
@@ -106,6 +107,7 @@ COVERAGE=0
 DEB=0
 INSTALL=1
 OS=Linux
+PACKAGE_RELEASE=
 RPM=0
 TESTS=1
 TGZ=0
@@ -162,6 +164,12 @@ while [[ $# -ne 0 ]]; do
         --packages|-p)
             TGZ=1
             ;;
+        --package-release)
+            PACKAGE_RELEASE=$2
+            build_arguments+=($1 $2)
+            shift 2
+            continue
+            ;;
         --release|-r)
             cmake_build_types+=(RelWithDebInfo)
             ;;
@@ -194,6 +202,11 @@ done
 
 cmake_arguments+=(-D BUILD_TESTING=$TESTS)
 
+if [[ -n "$PACKAGE_RELEASE" ]]; then
+    export DCGM_PACKAGE_RELEASE="$PACKAGE_RELEASE"
+    intodocker_arguments+=(--env "DCGM_PACKAGE_RELEASE=$DCGM_PACKAGE_RELEASE")
+fi
+
 if [[ ${DCGM_BUILD_INSIDE_DOCKER:-0} -eq 0 ]]; then
     if [[ $DCGM_SKIP_LFS_INSTALL -eq 0 ]]; then
         if ! git config --local --get filter.lfs.smudge > /dev/null; then

diff --git a/cmake/packaging.cmake b/cmake/packaging.cmake
@@ -16,6 +16,11 @@
 set(CPACK_PACKAGE_CONTACT "dcgm-support <dcgm-support@nvidia.com>")
 set(CPACK_PACKAGE_VENDOR "NVIDIA Corp.")
 
+set(DCGM_PACKAGE_RELEASE "1")
+if (DEFINED ENV{DCGM_PACKAGE_RELEASE} AND NOT "$ENV{DCGM_PACKAGE_RELEASE}" STREQUAL "")
+    set(DCGM_PACKAGE_RELEASE "$ENV{DCGM_PACKAGE_RELEASE}")
+endif()
+
 set(CPACK_PACKAGE_DESCRIPTION
 "This package is a component of DCGM, a userspace library and service that
 simplifies the administration of NVIDIA datacenter hardware in cluster and
@@ -122,7 +127,7 @@ if (CPACK_GENERATOR MATCHES "DEB")
     endif()
 
     set(CPACK_COMPONENTS_GROUPING "IGNORE")
-    set(CPACK_DEBIAN_PACKAGE_RELEASE "1")
+    set(CPACK_DEBIAN_PACKAGE_RELEASE "${DCGM_PACKAGE_RELEASE}")
     set(CPACK_DEBIAN_PACKAGE_CONTROL_STRICT_PERMISSION TRUE)
     set(CPACK_PRE_BUILD_SCRIPTS
         "${CMAKE_CURRENT_LIST_DIR}/cpack-deb-prebuild.cmake")
@@ -267,7 +272,7 @@ elseif(CPACK_GENERATOR MATCHES "RPM")
     set(CPACK_RPM_PACKAGE_AUTOREQ "no") # Removes dependency on libcuda.so.1
     set(CPACK_RPM_PACKAGE_DESCRIPTION "${CPACK_PACKAGE_DESCRIPTION_SUMMARY}")
     set(CPACK_RPM_PACKAGE_LICENSE "NVIDIA Proprietary")
-    set(CPACK_RPM_PACKAGE_RELEASE 1) # this is the package spec version, not the software version
+    set(CPACK_RPM_PACKAGE_RELEASE "${DCGM_PACKAGE_RELEASE}") # this is the package spec version, not the software version
     set(CPACK_RPM_SPEC_MORE_DEFINE "
 %define __strip ${CPACK_STRIP}
 %define __objdump ${CPACK_OBJDUMP}

diff --git a/dcgmbuild/container-images/common-host-software/scripts/urls.txt b/dcgmbuild/container-images/common-host-software/scripts/urls.txt
@@ -1,5 +1,5 @@
 ccache https://github.com/ccache/ccache/releases/download/v4.12.2/ccache-4.12.2-linux-x86_64.tar.xz d5aa5316d18bbb68ba332deca057e9f87e997f46316cb20beb4ef7e264f9181242d80b39629c59b92aff5fe0a1ce83bd35eb398a7f0353cff4ef0aa2730edeff
-clang https://apt.llvm.org/llvm.sh e8f246b75b2c2c7cb903e24c07257eea344aecf3f2a80c56896261ec61605eb223ad1795bfb9530dd8b4bef9b25ba11a6c32fdd6bdafbea34262f0ba97d01b08
+clang https://apt.llvm.org/llvm.sh cafe8efb95c27efd8bc4e7e936c18feb83b51bd73229f28b503cbeb132b99b90ea0d3324d84c67953482c5c15683c406db28e493ca0e88339945dbacdc8019d0
 cmake https://github.com/Kitware/CMake/releases/download/v4.1.1/cmake-4.1.1-linux-x86_64.sh c51ce071fff93f6825d2f2496d9adb3bd331c4ccbb7204278770fb2890973cbb4884ad766d084cdd2adbda96659bf69a694aca29d3e1cb0a99cf141abee8a9d3
 git-lfs https://github.com/git-lfs/git-lfs/releases/download/v2.13.3/git-lfs-linux-amd64-v2.13.3.tar.gz 110906644dd558705b996271066c18cc3e017035ceecf6dcea8a600691914513204f25a6d1549d20ca398d9bab78993d08ef66cc1744d1df8c74fadbdec965e7
 lcov https://github.com/linux-test-project/lcov/archive/v2.0.tar.gz 5d64b77154ab77d04607ad7bd18afa370ccce90d74a7a48ed8f2f506534ee856e9b2bd1271931797f6e79ee37b0c0f9b65d9dd2cfed99377b708679ca0cab772

diff --git a/dcgmlib/src/DcgmCacheManager.cpp b/dcgmlib/src/DcgmCacheManager.cpp
@@ -5838,9 +5838,6 @@ bool DcgmCacheManager::IsModulePushedFieldId(unsigned int fieldId)
         case DCGM_FI_DEV_NVLINK_COUNT_EFFECTIVE_BER_FLOAT:
         case DCGM_FI_DEV_NVLINK_COUNT_EFFECTIVE_ERRORS:
         case DCGM_FI_DEV_NVLINK_ECC_DATA_ERROR_COUNT_TOTAL:
-        case DCGM_FI_DEV_MEMORY_UNREPAIRABLE_FLAG:
-        case DCGM_FI_DEV_NVLINK_GET_STATE:
-        case DCGM_FI_DEV_NVLINK_PPCNT_IBPC_PORT_XMIT_WAIT:
         case DCGM_FI_DEV_C2C_LINK_ERROR_INTR:
         case DCGM_FI_DEV_C2C_LINK_ERROR_REPLAY:
         case DCGM_FI_DEV_C2C_LINK_ERROR_REPLAY_B2B:
@@ -5861,6 +5858,34 @@ bool DcgmCacheManager::IsModulePushedFieldId(unsigned int fieldId)
         case DCGM_FI_DEV_NVLINK_COUNT_FEC_HISTORY_13:
         case DCGM_FI_DEV_NVLINK_COUNT_FEC_HISTORY_14:
         case DCGM_FI_DEV_NVLINK_COUNT_FEC_HISTORY_15:
+        case DCGM_FI_DEV_CLOCKS_EVENT_REASON_SW_POWER_CAP_NS:
+        case DCGM_FI_DEV_CLOCKS_EVENT_REASON_SYNC_BOOST_NS:
+        case DCGM_FI_DEV_CLOCKS_EVENT_REASON_SW_THERM_SLOWDOWN_NS:
+        case DCGM_FI_DEV_CLOCKS_EVENT_REASON_HW_THERM_SLOWDOWN_NS:
+        case DCGM_FI_DEV_CLOCKS_EVENT_REASON_HW_POWER_BRAKE_SLOWDOWN_NS:
+        case DCGM_FI_DEV_PWR_SMOOTHING_ENABLED:
+        case DCGM_FI_DEV_PWR_SMOOTHING_PRIV_LVL:
+        case DCGM_FI_DEV_PWR_SMOOTHING_IMM_RAMP_DOWN_ENABLED:
+        case DCGM_FI_DEV_PWR_SMOOTHING_APPLIED_TMP_CEIL:
+        case DCGM_FI_DEV_PWR_SMOOTHING_APPLIED_TMP_FLOOR:
+        case DCGM_FI_DEV_PWR_SMOOTHING_MAX_PERCENT_TMP_FLOOR_SETTING:
+        case DCGM_FI_DEV_PWR_SMOOTHING_MIN_PERCENT_TMP_FLOOR_SETTING:
+        case DCGM_FI_DEV_PWR_SMOOTHING_HW_CIRCUITRY_PERCENT_LIFETIME_REMAINING:
+        case DCGM_FI_DEV_PWR_SMOOTHING_MAX_NUM_PRESET_PROFILES:
+        case DCGM_FI_DEV_PWR_SMOOTHING_PROFILE_PERCENT_TMP_FLOOR:
+        case DCGM_FI_DEV_PWR_SMOOTHING_PROFILE_RAMP_UP_RATE:
+        case DCGM_FI_DEV_PWR_SMOOTHING_PROFILE_RAMP_DOWN_RATE:
+        case DCGM_FI_DEV_PWR_SMOOTHING_PROFILE_RAMP_DOWN_HYST_VAL:
+        case DCGM_FI_DEV_PWR_SMOOTHING_ACTIVE_PRESET_PROFILE:
+        case DCGM_FI_DEV_PWR_SMOOTHING_ADMIN_OVERRIDE_PERCENT_TMP_FLOOR:
+        case DCGM_FI_DEV_PWR_SMOOTHING_ADMIN_OVERRIDE_RAMP_UP_RATE:
+        case DCGM_FI_DEV_PWR_SMOOTHING_ADMIN_OVERRIDE_RAMP_DOWN_RATE:
+        case DCGM_FI_DEV_PWR_SMOOTHING_ADMIN_OVERRIDE_RAMP_DOWN_HYST_VAL:
+        case DCGM_FI_DEV_PCIE_COUNT_CORRECTABLE_ERRORS:
+        case DCGM_FI_DEV_MEMORY_UNREPAIRABLE_FLAG:
+        case DCGM_FI_DEV_NVLINK_GET_STATE:
+        case DCGM_FI_DEV_NVLINK_PPCNT_IBPC_PORT_XMIT_WAIT:
+        case DCGM_FI_DEV_GET_GPU_RECOVERY_ACTION:
             return false;
         default:
             return true;