Merge pull request #303 from NVIDIA/ocp-b200-rhel9

[B200][OCP] pre-install nvlsm and infiniband pkgs

Author: tariq1890

rhel8/Dockerfile

@@ -81,6 +81,9 @@ RUN if [ "$DRIVER_TYPE" != "vgpu" ]; then \
 RUN if [ "$DRIVER_TYPE" != "vgpu" ] && [ "$DRIVER_BRANCH" -ge "550" ]; then \
     dnf install -y nvidia-imex-${DRIVER_BRANCH}-${DRIVER_VERSION}-1; fi
 
+RUN if [ "$DRIVER_TYPE" != "vgpu" ] && [ "$DRIVER_BRANCH" -ge "570" ]; then \
+    dnf install -y infiniband-diags nvlsm; fi
+
 COPY nvidia-driver /usr/local/bin
 COPY ocp_dtk_entrypoint /usr/local/bin
 COPY common.sh /usr/local/bin

rhel8/nvidia-driver

@@ -289,12 +289,6 @@ _ensure_nvlink5_prerequisites() (
         echo "waiting for the mlx5_core and ib_umad kernel modules to be loaded"
         sleep 10
     done
-
-    echo "Installing Infiniband packages"
-    dnf -q -y --releasever=${DNF_RELEASEVER} install libibverbs-utils infiniband-diags > /dev/null
-
-    echo "Installing the NVLink subnet manager..."
-    dnf -q -y --releasever=${DNF_RELEASEVER} install nvlsm > /dev/null
 )
 
 # For each kernel module configuration file mounted into the container,

rhel9/Dockerfile

@@ -75,6 +75,9 @@ RUN if [ "$DRIVER_TYPE" != "vgpu" ]; then \
 RUN if [ "$DRIVER_TYPE" != "vgpu" ] && [ "$DRIVER_BRANCH" -ge "550" ]; then \
     dnf install -y nvidia-imex-${DRIVER_BRANCH}-${DRIVER_VERSION}-1; fi
 
+RUN if [ "$DRIVER_TYPE" != "vgpu" ] && [ "$DRIVER_BRANCH" -ge "570" ]; then \
+    dnf install -y infiniband-diags nvlsm; fi
+
 COPY nvidia-driver /usr/local/bin
 COPY ocp_dtk_entrypoint /usr/local/bin
 COPY common.sh /usr/local/bin

rhel9/nvidia-driver

@@ -291,12 +291,6 @@ _ensure_nvlink5_prerequisites() (
         echo "waiting for the mlx5_core and ib_umad kernel modules to be loaded"
         sleep 10
     done
-
-    echo "Installing Infiniband packages..."
-    dnf -q -y --releasever=${DNF_RELEASEVER} install libibverbs-utils infiniband-diags > /dev/null
-
-    echo "Installing the NVLink subnet manager..."
-    dnf -q -y --releasever=${DNF_RELEASEVER} install nvlsm > /dev/null
 )
 
 # For each kernel module configuration file mounted into the container,