diff --git a/pkg/scheduler/algorithm/priorities/metadata.go b/pkg/scheduler/algorithm/priorities/metadata.go
index c80fc7d0358..9f34962f716 100644
--- a/pkg/scheduler/algorithm/priorities/metadata.go
+++ b/pkg/scheduler/algorithm/priorities/metadata.go
@@ -46,6 +46,7 @@ func NewPriorityMetadataFactory(serviceLister algorithm.ServiceLister, controlle
 // priorityMetadata is a type that is passed as metadata for priority functions
 type priorityMetadata struct {
 	nonZeroRequest          *schedulernodeinfo.Resource
+	podLimits               *schedulernodeinfo.Resource
 	podTolerations          []v1.Toleration
 	affinity                *v1.Affinity
 	podSelectors            []labels.Selector
@@ -62,6 +63,7 @@ func (pmf *PriorityMetadataFactory) PriorityMetadata(pod *v1.Pod, nodeNameToInfo
 	}
 	return &priorityMetadata{
 		nonZeroRequest:          getNonZeroRequests(pod),
+		podLimits:               getResourceLimits(pod),
 		podTolerations:          getAllTolerationPreferNoSchedule(pod.Spec.Tolerations),
 		affinity:                pod.Spec.Affinity,
 		podSelectors:            getSelectors(pod, pmf.serviceLister, pmf.controllerLister, pmf.replicaSetLister, pmf.statefulSetLister),
diff --git a/pkg/scheduler/algorithm/priorities/metadata_test.go b/pkg/scheduler/algorithm/priorities/metadata_test.go
index 9a992b5797f..bef089b37ff 100644
--- a/pkg/scheduler/algorithm/priorities/metadata_test.go
+++ b/pkg/scheduler/algorithm/priorities/metadata_test.go
@@ -38,6 +38,12 @@ func TestPriorityMetadata(t *testing.T) {
 	specifiedReqs.MilliCPU = 200
 	specifiedReqs.Memory = 2000
 
+	nonPodLimits := &schedulernodeinfo.Resource{}
+
+	specifiedPodLimits := &schedulernodeinfo.Resource{}
+	specifiedPodLimits.MilliCPU = 200
+	specifiedPodLimits.Memory = 2000
+
 	tolerations := []v1.Toleration{{
 		Key:      "foo",
 		Operator: v1.TolerationOpEqual,
@@ -104,6 +110,10 @@ func TestPriorityMetadata(t *testing.T) {
 					Image:           "image",
 					ImagePullPolicy: "Always",
 					Resources: v1.ResourceRequirements{
+						Limits: v1.ResourceList{
+							v1.ResourceCPU:    resource.MustParse("200m"),
+							v1.ResourceMemory: resource.MustParse("2000"),
+						},
 						Requests: v1.ResourceList{
 							v1.ResourceCPU:    resource.MustParse("200m"),
 							v1.ResourceMemory: resource.MustParse("2000"),
@@ -128,6 +138,7 @@ func TestPriorityMetadata(t *testing.T) {
 			pod: podWithTolerationsAndAffinity,
 			expected: &priorityMetadata{
 				nonZeroRequest: nonZeroReqs,
+				podLimits:      nonPodLimits,
 				podTolerations: tolerations,
 				affinity:       podAffinity,
 			},
@@ -137,6 +148,7 @@ func TestPriorityMetadata(t *testing.T) {
 			pod: podWithTolerationsAndRequests,
 			expected: &priorityMetadata{
 				nonZeroRequest: specifiedReqs,
+				podLimits:      nonPodLimits,
 				podTolerations: tolerations,
 				affinity:       nil,
 			},
@@ -146,6 +158,7 @@ func TestPriorityMetadata(t *testing.T) {
 			pod: podWithAffinityAndRequests,
 			expected: &priorityMetadata{
 				nonZeroRequest: specifiedReqs,
+				podLimits:      specifiedPodLimits,
 				podTolerations: nil,
 				affinity:       podAffinity,
 			},
diff --git a/pkg/scheduler/algorithm/priorities/resource_limits.go b/pkg/scheduler/algorithm/priorities/resource_limits.go
index 1344dc1eec4..8922007b1bb 100644
--- a/pkg/scheduler/algorithm/priorities/resource_limits.go
+++ b/pkg/scheduler/algorithm/priorities/resource_limits.go
@@ -42,7 +42,14 @@ func ResourceLimitsPriorityMap(pod *v1.Pod, meta interface{}, nodeInfo *schedule
 	allocatableResources := nodeInfo.AllocatableResource()
 
 	// compute pod limits
-	podLimits := getResourceLimits(pod)
+	var podLimits *schedulernodeinfo.Resource
+	if priorityMeta, ok := meta.(*priorityMetadata); ok && priorityMeta != nil {
+		// We were able to parse metadata, use podLimits from there.
+		podLimits = priorityMeta.podLimits
+	} else {
+		// We couldn't parse metadata - fallback to computing it.
+		podLimits = getResourceLimits(pod)
+	}
 
 	cpuScore := computeScore(podLimits.MilliCPU, allocatableResources.MilliCPU)
 	memScore := computeScore(podLimits.Memory, allocatableResources.Memory)
@@ -83,7 +90,6 @@ func computeScore(limit, allocatable int64) int64 {
 // The reason to create this new function is to be consistent with other
 // priority functions because most or perhaps all priority functions work
 // with schedulernodeinfo.Resource.
-// TODO: cache it as part of metadata passed to priority functions.
 func getResourceLimits(pod *v1.Pod) *schedulernodeinfo.Resource {
 	result := &schedulernodeinfo.Resource{}
 	for _, container := range pod.Spec.Containers {
diff --git a/pkg/scheduler/algorithm/priorities/resource_limits_test.go b/pkg/scheduler/algorithm/priorities/resource_limits_test.go
index 1e3782b014c..96bf356da32 100644
--- a/pkg/scheduler/algorithm/priorities/resource_limits_test.go
+++ b/pkg/scheduler/algorithm/priorities/resource_limits_test.go
@@ -27,7 +27,7 @@ import (
 	schedulernodeinfo "k8s.io/kubernetes/pkg/scheduler/nodeinfo"
 )
 
-func TestResourceLimistPriority(t *testing.T) {
+func TestResourceLimitsPriority(t *testing.T) {
 	noResources := v1.PodSpec{
 		Containers: []v1.Container{},
 	}
@@ -140,12 +140,22 @@ func TestResourceLimistPriority(t *testing.T) {
 	for _, test := range tests {
 		t.Run(test.name, func(t *testing.T) {
 			nodeNameToInfo := schedulernodeinfo.CreateNodeNameToInfoMap(nil, test.nodes)
-			list, err := priorityFunction(ResourceLimitsPriorityMap, nil, nil)(test.pod, nodeNameToInfo, test.nodes)
-			if err != nil {
-				t.Errorf("unexpected error: %v", err)
-			}
-			if !reflect.DeepEqual(test.expectedList, list) {
-				t.Errorf("expected %#v, got %#v", test.expectedList, list)
+
+			for _, hasMeta := range []bool{true, false} {
+				var metadata *priorityMetadata
+				if hasMeta {
+					metadata = &priorityMetadata{
+						podLimits: getResourceLimits(test.pod),
+					}
+				}
+
+				list, err := priorityFunction(ResourceLimitsPriorityMap, nil, metadata)(test.pod, nodeNameToInfo, test.nodes)
+				if err != nil {
+					t.Errorf("unexpected error: %v", err)
+				}
+				if !reflect.DeepEqual(test.expectedList, list) {
+					t.Errorf("expected %#v, got %#v", test.expectedList, list)
+				}
 			}
 		})
 	}