EKS GPUs - Disable GSP from userData script (#176)

truefoundry · Jan 16, 2024 · a053e92 · a053e92
1 parent 1d4ea2d
commit a053e92
Show file tree

Hide file tree

Showing 2 changed files with 12 additions and 1 deletion.
diff --git a/charts/tfy-karpenter-config/Chart.yaml b/charts/tfy-karpenter-config/Chart.yaml
@@ -2,4 +2,4 @@ apiVersion: v2
 name: tfy-karpenter-config
 description: "ArgoCD Applications for karpenter config"
 type: application
-version: 0.1.18-rc.6
+version: 0.1.18-rc.7
diff --git a/charts/tfy-karpenter-config/values.yaml b/charts/tfy-karpenter-config/values.yaml
@@ -227,6 +227,17 @@ karpenter:
           echo "CONTAINERD_VERSION is empty or not within the specified range."
       fi
 
+      rmmod nvidia_drm
+      rmmod nvidia_modeset
+      rmmod nvidia_uvm
+      rmmod nvidia
+      echo "Writing NVreg_EnableGpuFirmware=0 to /etc/modprobe.d/nvidia.conf"
+      echo "options nvidia NVreg_EnableGpuFirmware=0" | tee --append /etc/modprobe.d/nvidia.conf
+      echo "Writing NVreg_EnableGpuFirmware=0 to /etc/modprobe.d/nvidia-gsp.conf"
+      echo "options nvidia NVreg_EnableGpuFirmware=0" | tee --append /etc/modprobe.d/nvidia-gsp.conf
+      echo "Running dracut"
+      dracut -f
+
   ## Default Inferentia Node Template Config
   inferentiaDefaultNodeTemplate:
     ## @param karpenter.inferentiaDefaultNodeTemplate.rootVolumeSize Size for the root volume attached to node