diff --git a/deployments/gpu_tensorflow_test/deployment.yaml b/deployments/gpu_tensorflow_test/deployment.yaml
new file mode 100644
index 000000000..2a271c6bc
--- /dev/null
+++ b/deployments/gpu_tensorflow_test/deployment.yaml
@@ -0,0 +1,24 @@
+apiVersion: v1
+kind: Pod
+metadata:
+  name: training-pod
+spec:
+  restartPolicy: Never
+  containers:
+  - name: testcontainer
+    image: intel/intel-extension-for-tensorflow:1.1.0-gpu-flex
+    imagePullPolicy: IfNotPresent
+    command: ["/bin/sh", "-c"]
+    args: ["python /code/training.py"]
+    resources:
+      limits:
+        gpu.intel.com/i915: 1
+      requests:
+        gpu.intel.com/i915: 1
+    volumeMounts:
+    - mountPath: /code
+      name: code
+  volumes:
+  - configMap:
+      name: training-code
+    name: code
diff --git a/deployments/gpu_tensorflow_test/kustomization.yaml b/deployments/gpu_tensorflow_test/kustomization.yaml
new file mode 100644
index 000000000..66dcc11f2
--- /dev/null
+++ b/deployments/gpu_tensorflow_test/kustomization.yaml
@@ -0,0 +1,7 @@
+configMapGenerator:
+- name: training-code
+  files:
+  - training.py
+
+resources:
+  - deployment.yaml
diff --git a/deployments/gpu_tensorflow_test/training.py b/deployments/gpu_tensorflow_test/training.py
new file mode 100644
index 000000000..91879d23c
--- /dev/null
+++ b/deployments/gpu_tensorflow_test/training.py
@@ -0,0 +1,46 @@
+# original code from:
+# https://github.com/tensorflow/examples/blob/master/courses/udacity_intro_to_tensorflow_for_deep_learning/l02c01_celsius_to_fahrenheit.ipynb
+# this is slightly modified to run explicitly with XPU devices
+
+import tensorflow as tf
+import intel_extension_for_tensorflow as itex
+import numpy as np
+
+print("BACKENDS: ", str(itex.get_backend()))
+
+devs = tf.config.list_physical_devices('XPU')
+
+print(devs)
+
+if not devs:
+  raise Exception("No devices found")
+
+with tf.device("/xpu:0"):
+  celsius_q    = np.array([-40, -10,  0,  8, 15, 22,  38],  dtype=float)
+  fahrenheit_a = np.array([-40,  14, 32, 46, 59, 72, 100],  dtype=float)
+
+  model = tf.keras.Sequential([
+    tf.keras.layers.Dense(units=1, input_shape=[1])
+  ])
+
+  model.compile(loss='mean_squared_error',
+                optimizer=tf.keras.optimizers.Adam(0.1))
+
+  history = model.fit(celsius_q, fahrenheit_a, epochs=500, verbose=False)
+
+  print("model trained")
+
+  test = [100.0]
+  p = model.predict(test)
+
+  if len(p) != 1:
+    raise Exception("invalid result obj")
+
+  prediction = p[0]
+
+  if prediction >= 211 and prediction <= 213:
+    print("inference ok: %f" % prediction)
+  else:
+    raise Exception("bad prediction %f" % prediction)
+
+  print("SUCCESS")
diff --git a/test/e2e/gpu/gpu.go b/test/e2e/gpu/gpu.go
index f80b82733..d9ce24821 100644
--- a/test/e2e/gpu/gpu.go
+++ b/test/e2e/gpu/gpu.go
@@ -35,8 +35,10 @@ import (
 )
 
 const (
-	kustomizationYaml = "deployments/gpu_plugin/kustomization.yaml"
-	containerName     = "testcontainer"
+	kustomizationYaml   = "deployments/gpu_plugin/kustomization.yaml"
+	containerName       = "testcontainer"
+	tfKustomizationYaml = "deployments/gpu_tensorflow_test/kustomization.yaml"
+	tfPodName           = "training-pod"
 )
 
 func init() {
@@ -113,4 +115,50 @@ func describe() {
 
 		framework.Logf("found card and renderD from the log")
 	})
+
+	ginkgo.It("run some tensorflow code on GPU", func() {
+		ginkgo.By("deploying GPU plugin")
+		e2ekubectl.RunKubectlOrDie(f.Namespace.Name, "apply", "-k", filepath.Dir(kustomizationPath))
+
+		ginkgo.By("waiting for GPU plugin's availability")
+		_, err := e2epod.WaitForPodsWithLabelRunningReady(f.ClientSet, f.Namespace.Name,
+			labels.Set{"app": "intel-gpu-plugin"}.AsSelector(), 1 /* one replica */, 100*time.Second)
+		if err != nil {
+			e2edebug.DumpAllNamespaceInfo(f.ClientSet, f.Namespace.Name)
+			e2ekubectl.LogFailedContainers(f.ClientSet, f.Namespace.Name, framework.Logf)
+			framework.Failf("unable to wait for all pods to be running and ready: %v", err)
+		}
+
+		ginkgo.By("checking if the resource is allocatable")
+		if err = utils.WaitForNodesWithResource(f.ClientSet, "gpu.intel.com/i915", 30*time.Second); err != nil {
+			framework.Failf("unable to wait for nodes to have positive allocatable resource: %v", err)
+		}
+
+		kustomYaml, err := utils.LocateRepoFile(tfKustomizationYaml)
+		if err != nil {
+			framework.Failf("unable to locate %q: %v", tfKustomizationYaml, err)
+		}
+
+		ginkgo.By("submitting demo deployment")
+
+		e2ekubectl.RunKubectlOrDie(f.Namespace.Name, "apply", "-k", filepath.Dir(kustomYaml))
+
+		ginkgo.By("waiting the pod to finish")
+		e2epod.NewPodClient(f).WaitForFinish(tfPodName, 240*time.Second)
+
+		ginkgo.By("checking log output")
+		log, err := e2epod.GetPodLogs(f.ClientSet, f.Namespace.Name, tfPodName, containerName)
+
+		if err != nil {
+			framework.Logf("logs: %s", log)
+
+			framework.Failf("unable to get log from pod: %v", err)
+		}
+
+		if !strings.Contains(log, "SUCCESS") {
+			framework.Failf("tensorflow execution failed")
+		}
+
+		framework.Logf("tensorflow execution succeeded!")
+	})
 }