diff --git a/tests/k6/components/k8s.js b/tests/k6/components/k8s.js
index 4472c899e0..32982de322 100644
--- a/tests/k6/components/k8s.js
+++ b/tests/k6/components/k8s.js
@@ -37,17 +37,11 @@ function seldonObjExists(kind, name, ns) {
 }
 
 export function loadModel(modelName, data, awaitReady=true) {
-    // TODO: Update existing model with new CR definition.
-    // At the moment, if an object with the same name exists, it will not be
-    // re-loaded with different settings. This is because we get a k8s apply
-    // conflict caused by a FieldManager being set on `.spec.memory`
-    if(!seldonObjExists(seldonObjectType.MODEL, modelName, namespace)) {
-        kubeclient.apply(data)
-        let created = kubeclient.get(seldonObjectType.MODEL.description, modelName, namespace)
-        if ('uid' in created.metadata) {
-            if (awaitReady && schedulerClient != null) {
-                awaitStatus(modelName, "ModelAvailable")
-            }
+    kubeclient.apply(data)
+    let created = kubeclient.get(seldonObjectType.MODEL.description, modelName, namespace)
+    if ('uid' in created.metadata) {
+        if (awaitReady && schedulerClient != null) {
+            awaitStatus(modelName, "ModelAvailable")
         }
     }
 }
@@ -62,13 +56,11 @@ export function unloadModel(modelName, awaitReady=true) {
 }
 
 export function loadPipeline(pipelineName, data, awaitReady=true) {
-    if(!seldonObjExists(seldonObjectType.PIPELINE, pipelineName, namespace)) {
-        kubeclient.apply(data)
-        let created = kubeclient.get(seldonObjectType.PIPELINE.description, pipelineName, namespace)
-        if ('uid' in created.metadata) {
-            if (awaitReady && schedulerClient != null) {
-                awaitStatus(pipelineName, "PipelineReady")
-            }
+    kubeclient.apply(data)
+    let created = kubeclient.get(seldonObjectType.PIPELINE.description, pipelineName, namespace)
+    if ('uid' in created.metadata) {
+        if (awaitReady && schedulerClient != null) {
+            awaitStatus(pipelineName, "PipelineReady")
         }
     }
 }
@@ -83,13 +75,11 @@ export function unloadPipeline(pipelineName, awaitReady = true) {
 }
 
 export function loadExperiment(experimentName, data, awaitReady=true) {
-    if(!seldonObjExists(seldonObjectType.EXPERIMENT, experimentName, namespace)) {
-        kubeclient.apply(data)
-        let created = kubeclient.get(seldonObjectType.EXPERIMENT.description, experimentName, namespace)
-        if ('uid' in created.metadata) {
-            if (awaitReady && schedulerClient != null) {
-                awaitExperimentStart(experimentName)
-            }
+    kubeclient.apply(data)
+    let created = kubeclient.get(seldonObjectType.EXPERIMENT.description, experimentName, namespace)
+    if ('uid' in created.metadata) {
+        if (awaitReady && schedulerClient != null) {
+            awaitExperimentStart(experimentName)
         }
     }
 }
diff --git a/tests/k6/configs/k8s/base/k6.yaml b/tests/k6/configs/k8s/base/k6.yaml
index f005a361a6..d4adda8eb0 100644
--- a/tests/k6/configs/k8s/base/k6.yaml
+++ b/tests/k6/configs/k8s/base/k6.yaml
@@ -57,8 +57,13 @@ spec:
           value: "tfsimplea,pytorch-cifar10a,tfmnista,mlflow-winea,irisa"
         - name: MODEL_TYPE
           value: "tfsimple,pytorch_cifar10,tfmnist,mlflow_wine,iris"
+        # Specify MODEL_MEMORY_BYTES with suffixes (b, k, m, g) rather than
+        # numbers without units of measure. If supplying "naked numbers",
+        # the seldon operator will take care of converting the number for you
+        # but also take ownership of the field (as FieldManager), so the next
+        # time you run the scenario creating/updating of the model CR will fail.
         - name: MODEL_MEMORY_BYTES
-          value: "400000,8000000,43000000,200000,3000000"
+          value: "400k,8m,43m,200k,3m"
         - name: MAX_NUM_MODELS
           value: "800,100,25,100,100"
           # value: "0,0,25,100,100"