diff --git a/cluster/addons/device-plugins/nvidia-gpu/daemonset.yaml b/cluster/addons/device-plugins/nvidia-gpu/daemonset.yaml
index a5eaf1da76d..6b5edbf733d 100644
--- a/cluster/addons/device-plugins/nvidia-gpu/daemonset.yaml
+++ b/cluster/addons/device-plugins/nvidia-gpu/daemonset.yaml
@@ -11,7 +11,10 @@ spec:
     metadata:
       labels:
         k8s-app: nvidia-gpu-device-plugin
+      annotations:
+        scheduler.alpha.kubernetes.io/critical-pod: ''
     spec:
+      priorityClassName: system-node-critical
       affinity:
         nodeAffinity:
           requiredDuringSchedulingIgnoredDuringExecution:
@@ -34,7 +37,10 @@ spec:
         name: nvidia-gpu-device-plugin
         resources:
           requests:
-            cpu: 10m
+            cpu: 50m
+            memory: 10Mi
+          limits:
+            cpu: 50m
             memory: 10Mi
         securityContext:
           privileged: true
diff --git a/test/e2e/scheduling/nvidia-gpus.go b/test/e2e/scheduling/nvidia-gpus.go
index 785791e9e98..442f758d287 100644
--- a/test/e2e/scheduling/nvidia-gpus.go
+++ b/test/e2e/scheduling/nvidia-gpus.go
@@ -183,6 +183,11 @@ func testNvidiaGPUsOnCOS(f *framework.Framework) {
 
 	pods, err := framework.WaitForControlledPods(f.ClientSet, ds.Namespace, ds.Name, extensionsinternal.Kind("DaemonSet"))
 	framework.ExpectNoError(err, "getting pods controlled by the daemonset")
+	devicepluginPods, err := framework.WaitForControlledPods(f.ClientSet, "kube-system", "nvidia-gpu-device-plugin", extensionsinternal.Kind("DaemonSet"))
+	if err == nil {
+		framework.Logf("Adding deviceplugin addon pod.")
+		pods.Items = append(pods.Items, devicepluginPods.Items...)
+	}
 	framework.Logf("Starting ResourceUsageGather for the created DaemonSet pods.")
 	rsgather, err := framework.NewResourceUsageGatherer(f.ClientSet, framework.ResourceGathererOptions{false, false, 2 * time.Second, 2 * time.Second, true}, pods)
 	framework.ExpectNoError(err, "creating ResourceUsageGather for the daemonset pods")