Merge remote-tracking branch 'origin/main' into imex-support

Author: guptaNswati

.common-ci.yml

@@ -19,7 +19,7 @@ default:
       command: ["--experimental"]
 
 variables:
-  DRIVER_VERSIONS: 535.216.01 550.127.05
+  DRIVER_VERSIONS: 535.216.01 550.127.05 565.57.01
   BUILD_MULTI_ARCH_IMAGES: "true"
 
 stages:
@@ -73,13 +73,7 @@ trigger-pipeline:
 .driver-versions:
   parallel:
     matrix:
-      - DRIVER_VERSION: [535.216.01, 550.127.05]
-
-# Define the driver versions for jobs that can be run in parallel for ubuntu22.04
-.driver-versions-ubuntu22.04:
-  parallel:
-    matrix:
-      - DRIVER_VERSION: [535.216.01, 550.127.05]
+      - DRIVER_VERSION: [535.216.01, 550.127.05, 565.57.01]
 
 # Define the matrix of precompiled jobs that can be run in parallel for ubuntu22.04
 .driver-versions-precompiled-ubuntu22.04:
@@ -88,12 +82,6 @@ trigger-pipeline:
       - DRIVER_BRANCH: [535, 550]
         KERNEL_FLAVOR: [aws, azure, generic, nvidia, oracle]
 
-# Define the driver versions for jobs that can be run in parallel for rhel9
-.driver-versions-rhel9:
-  parallel:
-    matrix:
-      - DRIVER_VERSION: [535.216.01, 550.127.05]
-
 # Define the distribution targets
 .dist-ubuntu20.04:
   variables:
@@ -170,15 +158,15 @@ trigger-pipeline:
   # Perform for each DRIVER_VERSION
   extends:
     - .release-generic
-    - .driver-versions-ubuntu22.04
+    - .driver-versions
   rules:
     - if: $CI_PIPELINE_SOURCE != "schedule"
 
 .release-rhel9:
   # Perform for each DRIVER_VERSION
   extends:
     - .release-generic
-    - .driver-versions-rhel9
+    - .driver-versions
   rules:
     - if: $CI_PIPELINE_SOURCE != "schedule"
 

.github/workflows/image.yaml

@@ -36,7 +36,7 @@ jobs:
         driver:
           - 535.216.01
           - 550.127.05
-          - 560.35.03
+          - 565.57.01
         dist:
           - ubuntu20.04
           - ubuntu22.04

.gitlab-ci.yml

@@ -44,7 +44,7 @@ include:
 .image-build-ubuntu22.04:
   # Perform for each DRIVER_VERSION
   extends:
-    - .driver-versions-ubuntu22.04
+    - .driver-versions
     - .image-build-generic
   rules:
     - if: $CI_PIPELINE_SOURCE != "schedule"
@@ -53,7 +53,7 @@ include:
 .image-build-rhel9:
   # Perform for each DRIVER_VERSION
   extends:
-    - .driver-versions-rhel9
+    - .driver-versions
     - .image-build-generic
   rules:
     - if: $CI_PIPELINE_SOURCE != "schedule"

.nvidia-ci.yml

@@ -80,7 +80,7 @@ variables:
 .image-pull-ubuntu22.04:
   # Perform for each DRIVER_VERSION
   extends:
-    - .driver-versions-ubuntu22.04
+    - .driver-versions
     - .image-pull-generic
   rules:
     - if: $CI_PIPELINE_SOURCE == "schedule"
@@ -175,7 +175,7 @@ image-rhel8:
 .scan-ubuntu22.04:
   # Repeat for each DRIVER_VERSION
   extends:
-    - .driver-versions-ubuntu22.04
+    - .driver-versions
     - .scan-generic
   rules:
     - !reference [.scan-rules-common, rules]
@@ -276,7 +276,7 @@ release:ngc-ubuntu22.04:
   extends:
     - .release:ngc
     - .dist-ubuntu22.04
-    - .driver-versions-ubuntu22.04
+    - .driver-versions
 
 release:ngc-precompiled-ubuntu22.04:
   variables:
@@ -429,13 +429,13 @@ sign:ngc-ubuntu-rhel-rhcos:
     matrix:
     - SIGN_JOB_NAME: ["ubuntu"]
       VERSION: ["22.04"]
-      DRIVER_VERSION: ["535.216.01", "550.127.05"]
+      DRIVER_VERSION: ["535.216.01", "550.127.05", "565.57.01"]
     - SIGN_JOB_NAME: ["ubuntu"]
       VERSION: ["20.04"]
-      DRIVER_VERSION: ["535.216.01", "550.127.05"]
+      DRIVER_VERSION: ["535.216.01", "550.127.05", "565.57.01"]
     - SIGN_JOB_NAME: ["rhel"]
       VERSION: ["8.8", "8.10"]
-      DRIVER_VERSION: ["535.216.01", "550.127.05"]
+      DRIVER_VERSION: ["535.216.01", "550.127.05", "565.57.01"]
     - SIGN_JOB_NAME: ["rhcos"]
       VERSION: ["4.12","4.13","4.14","4.15", "4.16", "4.17"]
-      DRIVER_VERSION: ["535.216.01", "550.127.05"]
+      DRIVER_VERSION: ["535.216.01", "550.127.05", "565.57.01"]

rhel8/Dockerfile

@@ -76,7 +76,7 @@ RUN if [ "$DRIVER_TYPE" != "vgpu" ] && [ "$TARGETARCH" != "arm64" ]; then \
       fmPackage=nvidia-fabricmanager-${DRIVER_BRANCH}-${DRIVER_VERSION}-1; \
     fi; \
     nscqPackage=libnvidia-nscq-${DRIVER_BRANCH}-${DRIVER_VERSION}-1; \
-    dnf module enable -y nvidia-driver:${DRIVER_BRANCH} && \
+    dnf module enable -y nvidia-driver:${DRIVER_BRANCH}-dkms && \
     dnf install -y ${fmPackage} ${nscqPackage}; fi
 
 COPY nvidia-driver /usr/local/bin

rhel9/Dockerfile

@@ -70,7 +70,7 @@ RUN if [ "$DRIVER_TYPE" != "vgpu" ]; then \
 RUN if [ "$DRIVER_TYPE" != "vgpu" ] && [ "$TARGETARCH" != "arm64" ]; then \
     versionArray=(${DRIVER_VERSION//./ }); \
     DRIVER_BRANCH=${versionArray[0]}; \
-    dnf module enable -y nvidia-driver:${DRIVER_BRANCH} && \
+    dnf module enable -y nvidia-driver:${DRIVER_BRANCH}-dkms && \
     dnf install -y nvidia-fabric-manager-${DRIVER_VERSION}-1 libnvidia-nscq-${DRIVER_BRANCH}-${DRIVER_VERSION}-1; fi
 
 COPY nvidia-driver /usr/local/bin

versions.mk

@@ -13,8 +13,6 @@
 # limitations under the License.
 
 # DRIVER_VERSIONS contains latest version in all active datacenter branches
-DRIVER_VERSIONS ?= 535.216.01 550.127.05
-
-CUDA_VERSION := 12.6.0
+DRIVER_VERSIONS ?= 535.216.01 550.127.05 565.57.01
 
 GOLANG_VERSION := 1.22.8