diff --git a/shared/source/memory_manager/unified_memory_manager.cpp b/shared/source/memory_manager/unified_memory_manager.cpp
index 104a991464..0d1c0041f4 100644
--- a/shared/source/memory_manager/unified_memory_manager.cpp
+++ b/shared/source/memory_manager/unified_memory_manager.cpp
@@ -298,7 +298,7 @@ void *SVMAllocsManager::createUnifiedMemoryAllocation(size_t size,
     if (memoryProperties.memoryType == InternalMemoryType::deviceUnifiedMemory) {
         unifiedMemoryProperties.flags.isUSMDeviceAllocation = true;
         if (this->usmDeviceAllocationsCacheEnabled &&
-            false == memoryProperties.needZeroedOutAllocation) {
+            false == memoryProperties.isInternalAllocation) {
             void *allocationFromCache = this->usmDeviceAllocationsCache.get(size, memoryProperties, this);
             if (allocationFromCache) {
                 return allocationFromCache;
@@ -332,6 +332,7 @@ void *SVMAllocsManager::createUnifiedMemoryAllocation(size_t size,
     allocData.allocationFlagsProperty = memoryProperties.allocationFlags;
     allocData.device = memoryProperties.device;
     allocData.setAllocId(++this->allocationsCounter);
+    allocData.isInternalAllocation = memoryProperties.isInternalAllocation;
 
     auto retPtr = reinterpret_cast<void *>(unifiedMemoryAllocation->getGpuAddress());
     insertSVMAlloc(retPtr, allocData);
@@ -445,8 +446,9 @@ bool SVMAllocsManager::freeSVMAlloc(void *ptr, bool blocking) {
     SvmAllocationData *svmData = getSVMAlloc(ptr);
     if (svmData) {
         if (InternalMemoryType::deviceUnifiedMemory == svmData->memoryType &&
+            false == svmData->isInternalAllocation &&
             this->usmDeviceAllocationsCacheEnabled) {
-            if (this->usmDeviceAllocationsCache.insert(svmData->size, ptr)) {
+            if (this->usmDeviceAllocationsCache.insert(svmData->gpuAllocations.getDefaultGraphicsAllocation()->getUnderlyingBufferSize(), ptr)) {
                 return true;
             }
         }
diff --git a/shared/source/memory_manager/unified_memory_manager.h b/shared/source/memory_manager/unified_memory_manager.h
index c8de902c15..abce9ebefd 100644
--- a/shared/source/memory_manager/unified_memory_manager.h
+++ b/shared/source/memory_manager/unified_memory_manager.h
@@ -41,6 +41,7 @@ struct SvmAllocationData {
         this->allocId = svmAllocData.allocId;
         this->pageSizeForAlignment = svmAllocData.pageSizeForAlignment;
         this->isImportedAllocation = svmAllocData.isImportedAllocation;
+        this->isInternalAllocation = svmAllocData.isInternalAllocation;
         for (auto allocation : svmAllocData.gpuAllocations.getGraphicsAllocations()) {
             if (allocation) {
                 this->gpuAllocations.addAllocation(allocation);
@@ -63,6 +64,7 @@ struct SvmAllocationData {
         allocId = id;
     }
     bool mappedAllocData = false;
+    bool isInternalAllocation = false;
 
     uint32_t getAllocId() const {
         return allocId;
@@ -138,7 +140,7 @@ class SVMAllocsManager {
         const RootDeviceIndicesContainer &rootDeviceIndices;
         const std::map<uint32_t, DeviceBitfield> &subdeviceBitfields;
         AllocationType requestedAllocationType = AllocationType::unknown;
-        bool needZeroedOutAllocation = false;
+        bool isInternalAllocation = false;
     };
 
     struct SvmCacheAllocationInfo {
diff --git a/shared/source/program/program_initialization.cpp b/shared/source/program/program_initialization.cpp
index b42f2624a8..81e9665c46 100644
--- a/shared/source/program/program_initialization.cpp
+++ b/shared/source/program/program_initialization.cpp
@@ -38,7 +38,7 @@ GraphicsAllocation *allocateGlobalsSurface(NEO::SVMAllocsManager *const svmAlloc
         NEO::SVMAllocsManager::UnifiedMemoryProperties unifiedMemoryProperties(InternalMemoryType::deviceUnifiedMemory, 1, rootDeviceIndices, subDeviceBitfields);
         unifiedMemoryProperties.device = &device;
         unifiedMemoryProperties.requestedAllocationType = allocationType;
-        unifiedMemoryProperties.needZeroedOutAllocation = true;
+        unifiedMemoryProperties.isInternalAllocation = true;
         auto ptr = svmAllocManager->createUnifiedMemoryAllocation(totalSize, unifiedMemoryProperties);
         DEBUG_BREAK_IF(ptr == nullptr);
         if (ptr == nullptr) {
diff --git a/shared/test/common/mocks/mock_svm_manager.h b/shared/test/common/mocks/mock_svm_manager.h
index 3ee522d43e..640a0f4a26 100644
--- a/shared/test/common/mocks/mock_svm_manager.h
+++ b/shared/test/common/mocks/mock_svm_manager.h
@@ -34,7 +34,7 @@ struct MockSVMAllocsManager : public SVMAllocsManager {
     bool prefetchMemoryCalled = false;
 
     void *createUnifiedMemoryAllocation(size_t size, const UnifiedMemoryProperties &memoryProperties) override {
-        requestedZeroedOutAllocation = memoryProperties.needZeroedOutAllocation;
+        requestedZeroedOutAllocation = memoryProperties.isInternalAllocation;
         return SVMAllocsManager::createUnifiedMemoryAllocation(size, memoryProperties);
     }
     bool requestedZeroedOutAllocation = false;
diff --git a/shared/test/unit_test/memory_manager/unified_memory_manager_cache_tests.cpp b/shared/test/unit_test/memory_manager/unified_memory_manager_cache_tests.cpp
index 8885ba0f13..2d19e23a49 100644
--- a/shared/test/unit_test/memory_manager/unified_memory_manager_cache_tests.cpp
+++ b/shared/test/unit_test/memory_manager/unified_memory_manager_cache_tests.cpp
@@ -291,11 +291,8 @@ TEST_F(SvmDeviceAllocationCacheTest, givenAllocationsWithDifferentSizesWhenAlloc
     auto testDataset = std::vector<SvmDeviceAllocationCacheSimpleTestDataType>(
         {
             {(allocationSizeBasis << 0), nullptr},
-            {(allocationSizeBasis << 0) + 1, nullptr},
             {(allocationSizeBasis << 1), nullptr},
-            {(allocationSizeBasis << 1) + 1, nullptr},
             {(allocationSizeBasis << 2), nullptr},
-            {(allocationSizeBasis << 2) + 1, nullptr},
         });
 
     SVMAllocsManager::UnifiedMemoryProperties unifiedMemoryProperties(InternalMemoryType::deviceUnifiedMemory, 1, rootDeviceIndices, deviceBitfields);
@@ -519,7 +516,7 @@ TEST_F(SvmDeviceAllocationCacheTest, givenDeviceOutOfMemoryWhenAllocatingThenCac
     ASSERT_EQ(svmManager->usmDeviceAllocationsCache.allocations.size(), 0u);
 }
 
-TEST_F(SvmDeviceAllocationCacheTest, givenAllocationWithNeedZeroedOutAllocationWhenAllocatingAfterFreeThenDoNotReuseAllocation) {
+TEST_F(SvmDeviceAllocationCacheTest, givenAllocationWithIsInternalAllocationSetWhenAllocatingAfterFreeThenDoNotReuseAllocation) {
     std::unique_ptr<UltDeviceFactory> deviceFactory(new UltDeviceFactory(1, 1));
     RootDeviceIndicesContainer rootDeviceIndices = {mockRootDeviceIndex};
     std::map<uint32_t, DeviceBitfield> deviceBitfields{{mockRootDeviceIndex, mockDeviceBitfield}};
@@ -538,13 +535,15 @@ TEST_F(SvmDeviceAllocationCacheTest, givenAllocationWithNeedZeroedOutAllocationW
     svmManager->freeSVMAlloc(allocation);
     EXPECT_EQ(svmManager->usmDeviceAllocationsCache.allocations.size(), 1u);
 
-    unifiedMemoryProperties.needZeroedOutAllocation = true;
+    unifiedMemoryProperties.isInternalAllocation = true;
     auto testedAllocation = svmManager->createUnifiedMemoryAllocation(10u, unifiedMemoryProperties);
     EXPECT_EQ(svmManager->usmDeviceAllocationsCache.allocations.size(), 1u);
     auto svmData = svmManager->getSVMAlloc(testedAllocation);
     EXPECT_NE(nullptr, svmData);
+    EXPECT_TRUE(svmData->isInternalAllocation);
 
     svmManager->freeSVMAlloc(testedAllocation);
+    EXPECT_EQ(svmManager->usmDeviceAllocationsCache.allocations.size(), 1u);
 
     svmManager->trimUSMDeviceAllocCache();
 }
diff --git a/shared/test/unit_test/program/program_initialization_tests.cpp b/shared/test/unit_test/program/program_initialization_tests.cpp
index cd3bc49b5f..4db761a568 100644
--- a/shared/test/unit_test/program/program_initialization_tests.cpp
+++ b/shared/test/unit_test/program/program_initialization_tests.cpp
@@ -90,7 +90,9 @@ TEST(AllocateGlobalSurfaceTest, GivenSvmAllocsManagerWhenGlobalsAreExportedThenM
     EXPECT_EQ(0, memcmp(alloc->getUnderlyingBuffer(), initData.data(), initData.size()));
     ASSERT_NE(nullptr, svmAllocsManager.getSVMAlloc(reinterpret_cast<void *>(static_cast<uintptr_t>(alloc->getGpuAddress()))));
     EXPECT_TRUE(alloc->isMemObjectsAllocationWithWritableFlags());
-    EXPECT_EQ(InternalMemoryType::deviceUnifiedMemory, svmAllocsManager.getSVMAlloc(reinterpret_cast<void *>(alloc->getGpuAddress()))->memoryType);
+    auto svmData = svmAllocsManager.getSVMAlloc(reinterpret_cast<void *>(alloc->getGpuAddress()));
+    EXPECT_EQ(InternalMemoryType::deviceUnifiedMemory, svmData->memoryType);
+    EXPECT_TRUE(svmData->isInternalAllocation);
     EXPECT_EQ(AllocationType::constantSurface, alloc->getAllocationType());
     EXPECT_FALSE(alloc->getDefaultGmm()->resourceParams.Flags.Info.NotLockable);
     EXPECT_TRUE(svmAllocsManager.requestedZeroedOutAllocation);