Set CUDA12 as default in GPU packages

tools/ci_build/github/azure-pipelines/linux-gpu-ci-pipeline.yml

@@ -30,7 +30,7 @@ parameters:
   - name: CudaVersion
     displayName: CUDA version
     type: string
-    default: '11.8'
+    default: '12.2'
     values:
       - 11.8
       - 12.2

tools/ci_build/github/azure-pipelines/linux-gpu-tensorrt-ci-pipeline.yml

@@ -30,7 +30,7 @@ parameters:
   - name: CudaVersion
     displayName: CUDA version
     type: string
-    default: '11.8'
+    default: '12.2'
     values:
       - 11.8
       - 12.2

tools/ci_build/github/azure-pipelines/linux-gpu-tensorrt-daily-perf-pipeline.yml

@@ -8,7 +8,7 @@ parameters:
 - name: TrtVersion
   displayName: TensorRT Version
   type: string
-  default: 10.0.cuda_11_8_cudnn_8
+  default: 10.0.cuda_12_4_cudnn_9
   values:
   - 8.4.cuda_11_6_cudnn_8
   - 8.5.cuda_11_8_cudnn_8

tools/ci_build/github/azure-pipelines/nuget-cuda-publishing-pipeline.yml

@@ -6,6 +6,7 @@ resources:
       branches:
         include:
         - main
+        - rel-*
     branch: main
 
 parameters:
@@ -16,15 +17,15 @@ parameters:
 variables:
   - name: ArtifactFeed
     ${{ if eq(parameters.isReleaseBuild, false) }}:
-      value: ort-cuda-12-nightly
+      value: ORT-Nightly
     ${{ else }}:
       value: onnxruntime-cuda-12
 
 stages:
-- template: stages/nuget-cuda-publishing-stage.yml
-  parameters:
-    artifact_feed: $(ArtifactFeed)
+  - template: stages/nuget-cuda-publishing-stage.yml
+    parameters:
+      artifact_feed: $(ArtifactFeed)
 
-- template: stages/java-cuda-publishing-stage.yml
-  parameters:
-    artifact_feed: $(ArtifactFeed)
+  - template: stages/java-cuda-publishing-stage.yml
+    parameters:
+      artifact_feed: $(ArtifactFeed)

tools/ci_build/github/azure-pipelines/publish-nuget.yml

@@ -9,10 +9,22 @@ resources:
         - rel-*
     branch: main
 
+parameters:
+  - name: isReleaseBuild
+    type: boolean
+    default: false
+
+variables:
+  - name: ArtifactFeed
+    ${{ if eq(parameters.isReleaseBuild, false) }}:
+      value: ort-cuda-11-nightly
+    ${{ else }}:
+      value: onnxruntime-cuda-11
+
 stages:
   - template: templates/publish-nuget-steps.yml
     parameters:
-      stage_name: 'Publish_NuGet_Packag_And_Report'
+      stage_name: 'Publish_NuGet_Package_And_Report'
       include_cpu_ep: true
       download_artifacts_steps:
         - download: build
@@ -25,7 +37,11 @@ stages:
           artifact: 'drop-signed-nuget-Training-CPU'
         - script: move "$(Pipeline.Workspace)\build\drop-signed-nuget-Training-CPU\*" $(Build.BinariesDirectory)\nuget-artifact\final-package
 
-        - download: build
-          displayName: 'Download Pipeline Artifact - Signed NuGet Package'
-          artifact: 'drop-signed-nuget-GPU'
-        - script: move "$(Pipeline.Workspace)\build\drop-signed-nuget-GPU\*" $(Build.BinariesDirectory)\nuget-artifact\final-package
+  # Publish CUDA 11 Nuget/Java pkgs to ADO feed
+  - template: stages/nuget-cuda-publishing-stage.yml
+    parameters:
+      artifact_feed: $(ArtifactFeed)
+
+  - template: stages/java-cuda-publishing-stage.yml
+    parameters:
+      artifact_feed: $(ArtifactFeed)

tools/ci_build/github/azure-pipelines/py-cuda-publishing-pipeline.yml

@@ -16,7 +16,7 @@ parameters:
 variables:
   - name: ArtifactFeed
     ${{ if eq(parameters.isReleaseBuild, false) }}:
-      value: ort-cuda-12-nightly
+      value: ORT-Nightly
     ${{ else }}:
       value: onnxruntime-cuda-12
 

tools/ci_build/github/azure-pipelines/stages/java-cuda-publishing-stage.yml

@@ -8,7 +8,7 @@ stages:
   jobs:
   - job: JAR_Publishing_GPU
     #TD-DO: figure out a way to package nightly jar. Currently Java version are set from VERSION_NUMBER file
-    condition: ${{ eq(parameters.artifact_feed, 'onnxruntime-cuda-12') }}
+    condition: ${{ or(eq(parameters.artifact_feed, 'onnxruntime-cuda-11'), eq(parameters.artifact_feed, 'onnxruntime-cuda-12')) }}
     workspace:
       clean: all
     pool: 'onnxruntime-Win-CPU-2022'

tools/ci_build/github/azure-pipelines/templates/jobs/download_win_gpu_library.yml

@@ -7,7 +7,7 @@ parameters:
     default: false
   - name: CudaVersion
     type: string
-    default: '11.8'
+    default: '12.2'
     values:
       - 11.8
       - 12.2

tools/ci_build/github/azure-pipelines/templates/jobs/set-winenv.yml

@@ -9,10 +9,10 @@ parameters:
     default: false
   - name: PrimaryCUDAVersion
     type: string
-    default: '11.8'
+    default: '12.2'
   - name: SecondaryCUDAVersion
     type: string
-    default: '12.2'
+    default: '11.8'
 
 steps:
   - ${{ if eq(parameters.DownloadCUDA, 'true') }}:

tools/ci_build/github/azure-pipelines/win-gpu-ci-pipeline.yml

@@ -28,6 +28,13 @@ pr:
 #### end trigger ####
 
 parameters:
+- name: CudaVersion
+  displayName: CUDA version
+  type: string
+  default: '12.2'
+  values:
+    - 11.8
+    - 12.2
 - name: RunOnnxRuntimeTests
   displayName: Run Tests?
   type: boolean
@@ -43,7 +50,7 @@ stages:
         EnvSetupScript: setup_env_cuda.bat
         buildArch: x64
         additionalBuildFlags: >-
-          --enable_pybind --build_java --build_nodejs --use_cuda --cuda_home="$(Agent.TempDirectory)\v11.8"
+          --enable_pybind --build_java --build_nodejs --use_cuda --cuda_home="$(Agent.TempDirectory)\v${{ parameters.CudaVersion }}"
           --enable_cuda_profiling --enable_transformers_tool_test
           --cmake_extra_defines CMAKE_CUDA_ARCHITECTURES=86
           --cmake_extra_defines onnxruntime_BUILD_UNIT_TESTS=ON
@@ -65,7 +72,7 @@ stages:
         EnvSetupScript: setup_env_cuda.bat
         buildArch: x64
         additionalBuildFlags: >-
-          --enable_pybind --enable_training --use_cuda --cuda_home="$(Agent.TempDirectory)\v11.8"
+          --enable_pybind --enable_training --use_cuda --cuda_home="$(Agent.TempDirectory)\v${{ parameters.CudaVersion }}"
           --skip_onnx_tests
           --cmake_extra_defines CMAKE_CUDA_ARCHITECTURES=86
         msbuildPlatform: x64
@@ -105,7 +112,7 @@ stages:
         # note: need to specify `--gen_doc` when creating the build config so it has to be in additionalBuildFlags
         additionalBuildFlags: >-
           --gen_doc validate --skip_tests --enable_pybind --use_dml --use_cuda
-          --cuda_home="$(Agent.TempDirectory)\v11.8"
+          --cuda_home="$(Agent.TempDirectory)\v${{ parameters.CudaVersion }}"
           --cmake_extra_defines CMAKE_CUDA_ARCHITECTURES=86
           --cmake_extra_defines onnxruntime_BUILD_UNIT_TESTS=OFF
         msbuildPlatform: x64

tools/ci_build/github/azure-pipelines/win-gpu-tensorrt-ci-pipeline.yml

@@ -26,6 +26,21 @@ pr:
     - 'js/web'
     - 'onnxruntime/core/providers/js'
 #### end trigger ####
+parameters:
+- name: CudaVersion
+  displayName: CUDA version
+  type: string
+  default: '12.2'
+  values:
+    - 11.8
+    - 12.2
+
+variables:
+  - name: win_trt_folder
+    ${{ if eq(parameters.CudaVersion, '11.8') }}:
+      value: TensorRT-10.0.1.6.Windows10.x86_64.cuda-11.8
+    ${{ if eq(parameters.CudaVersion, '12.2') }}:
+      value: TensorRT-10.0.1.6.Windows10.x86_64.cuda-12.4
 
 jobs:
 - job: 'build'
@@ -55,7 +70,7 @@ jobs:
       WithCache: True
       Today: $(TODAY)
       AdditionalKey: "gpu-tensorrt | RelWithDebInfo"
-      BuildPyArguments: '--config RelWithDebInfo --parallel --use_binskim_compliant_compile_flags --build_dir $(Build.BinariesDirectory) --skip_submodule_sync --build_shared_lib --update --cmake_generator "Visual Studio 17 2022" --build_wheel --enable_onnx_tests --use_tensorrt --tensorrt_home="$(Agent.TempDirectory)\TensorRT-10.0.1.6.Windows10.x86_64.cuda-11.8" --cuda_home="$(Agent.TempDirectory)\v11.8" --cmake_extra_defines CMAKE_CUDA_ARCHITECTURES=86'
+      BuildPyArguments: '--config RelWithDebInfo --parallel --use_binskim_compliant_compile_flags --build_dir $(Build.BinariesDirectory) --skip_submodule_sync --build_shared_lib --update --cmake_generator "Visual Studio 17 2022" --build_wheel --enable_onnx_tests --use_tensorrt --tensorrt_home="$(Agent.TempDirectory)\${{ variables.win_trt_folder }}" --cuda_home="$(Agent.TempDirectory)\v${{ parameters.CudaVersion }}" --cmake_extra_defines CMAKE_CUDA_ARCHITECTURES=86'
       MsbuildArguments: $(MsbuildArguments)
       BuildArch: 'x64'
       Platform: 'x64'
@@ -75,7 +90,7 @@ jobs:
      del wheel_filename_file
      python.exe -m pip install -q --upgrade %WHEEL_FILENAME%
      set PATH=$(Build.BinariesDirectory)\RelWithDebInfo\RelWithDebInfo;%PATH%
-     python $(Build.SourcesDirectory)\tools\ci_build\build.py --config RelWithDebInfo --use_binskim_compliant_compile_flags --build_dir $(Build.BinariesDirectory) --skip_submodule_sync --build_shared_lib --test --cmake_generator "Visual Studio 17 2022" --build_wheel --enable_onnx_tests --use_tensorrt --tensorrt_home="$(Agent.TempDirectory)\TensorRT-10.0.1.6.Windows10.x86_64.cuda-11.8"  --cuda_home="$(Agent.TempDirectory)\v11.8" --cmake_extra_defines CMAKE_CUDA_ARCHITECTURES=75
+     python $(Build.SourcesDirectory)\tools\ci_build\build.py --config RelWithDebInfo --use_binskim_compliant_compile_flags --build_dir $(Build.BinariesDirectory) --skip_submodule_sync --build_shared_lib --test --cmake_generator "Visual Studio 17 2022" --build_wheel --enable_onnx_tests --use_tensorrt --tensorrt_home="$(Agent.TempDirectory)\${{ variables.win_trt_folder }}"  --cuda_home="$(Agent.TempDirectory)\v${{ parameters.CudaVersion }}" --cmake_extra_defines CMAKE_CUDA_ARCHITECTURES=86
 
     workingDirectory: '$(Build.BinariesDirectory)\RelWithDebInfo\RelWithDebInfo'
     displayName: 'Run tests'

tools/ci_build/github/linux/docker/Dockerfile.manylinux2_28_cuda

@@ -2,7 +2,7 @@
 # Please overwrite BASEIMAGE, TRT_VERSION and other arguments with
 # --docker-build-args ' --build-arg BASEIMAGE=other_base_image --build-arg TRT_VERSION=other_trt_version etc...'
 # for other cuda version and TRT version
-ARG BASEIMAGE=nvidia/cuda:11.8.0-cudnn8-devel-ubi8
+ARG BASEIMAGE=nvidia/cuda:12.5.1-cudnn-devel-ubi8
 
 FROM $BASEIMAGE
 ARG TRT_VERSION

tools/ci_build/github/linux/docker/Dockerfile.package_ubi8_cuda_tensorrt10_0

@@ -5,8 +5,8 @@
 # Dockerfile to Test ONNX Runtime on UBI8 with TensorRT 10.0 and CUDA 11.8 by default
 
 # Build base image with required system packages
-ARG BASEIMAGE=nvidia/cuda:11.8.0-cudnn8-devel-ubi8
-ARG TRT_VERSION=10.0.1.6-1.cuda11.8
+ARG BASEIMAGE=nvidia/cuda:12.5.1-cudnn-devel-ubi8
+ARG TRT_VERSION=10.0.1.6-1.cuda12.4
 FROM $BASEIMAGE AS base
 ARG TRT_VERSION
 ENV PATH /opt/python/cp38-cp38/bin:/usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/src/tensorrt/bin:${PATH}

tools/ci_build/github/linux/docker/Dockerfile.ubuntu_cuda12_4_tensorrt10_0

@@ -38,7 +38,22 @@ RUN version="10.0.1.6-1+cuda12.4" &&\
     apt-key adv --fetch-keys https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2004/x86_64/7fa2af80.pub &&\
     apt-get update &&\
     apt-get install -y \
-    tensorrt=${version}
+    libnvinfer-dev=${version} \
+    libnvinfer-dispatch-dev=${version} \
+    libnvinfer-dispatch10=${version} \
+    libnvinfer-headers-dev=${version} \
+    libnvinfer-headers-plugin-dev=${version} \
+    libnvinfer-lean-dev=${version} \
+    libnvinfer-lean10=${version} \
+    libnvinfer-plugin-dev=${version} \
+    libnvinfer-plugin10=${version} \
+    libnvinfer-vc-plugin-dev=${version} \
+    libnvinfer-vc-plugin10=${version} \
+    libnvinfer10=${version} \
+    libnvonnxparsers-dev=${version} \
+    libnvonnxparsers10=${version} \
+    tensorrt-dev=${version} \
+    libnvinfer-bin=${version}
 
 # Compile trtexec if not installed
 RUN if [ ! -d /usr/src/tensorrt/bin ] || [ ! -f /usr/src/tensorrt/bin/trtexec ]; then \