diff --git a/rocclr/device/pal/palresource.cpp b/rocclr/device/pal/palresource.cpp
index f8b3d4c6f5..00a2a017f4 100644
--- a/rocclr/device/pal/palresource.cpp
+++ b/rocclr/device/pal/palresource.cpp
@@ -1425,7 +1425,9 @@ bool Resource::partialMemCopyTo(VirtualGPU& gpu, const amd::Coord3D& srcOrigin,
     }
   }
 
-  bool cp_dma = dev().settings().disableSdma_;
+  bool cp_dma = dev().settings().disableSdma_ ||
+      (!enableCopyRect && desc().buffer_ && dstResource.desc().buffer_ &&
+       (size[0] < dev().settings().cpDmaCopySizeMax_));
   if (cp_dma) {
     // Make sure compute is done before CP DMA start
     gpu.addBarrier(RgpSqqtBarrierReason::MemDependency);
diff --git a/rocclr/device/pal/palsettings.cpp b/rocclr/device/pal/palsettings.cpp
index 20054df95b..dca774e22d 100644
--- a/rocclr/device/pal/palsettings.cpp
+++ b/rocclr/device/pal/palsettings.cpp
@@ -171,6 +171,7 @@ Settings::Settings() {
   mallPolicy_ = 0;
   alwaysResident_ = amd::IS_HIP ? true : false;
   prepinnedMinSize_ = 0;
+  cpDmaCopySizeMax_ = GPU_CP_DMA_COPY_SIZE * Ki;
 }
 
 bool Settings::create(const Pal::DeviceProperties& palProp,
diff --git a/rocclr/device/pal/palsettings.hpp b/rocclr/device/pal/palsettings.hpp
index c682dc3d3b..66984622ee 100644
--- a/rocclr/device/pal/palsettings.hpp
+++ b/rocclr/device/pal/palsettings.hpp
@@ -107,6 +107,7 @@ class Settings : public device::Settings {
   size_t stagedXferSize_;        //!< Staged buffer size
   size_t pinnedXferSize_;        //!< Pinned buffer size for transfer
   size_t pinnedMinXferSize_;     //!< Minimal buffer size for pinned transfer
+  size_t cpDmaCopySizeMax_;      //!< Threshold for CP DMA path in copy
   size_t resourceCacheSize_;     //!< Resource cache size in MB
   size_t numMemDependencies_;    //!< The array size for memory dependencies tracking
   uint64_t maxAllocSize_;        //!< Maximum single allocation size
diff --git a/rocclr/utils/flags.hpp b/rocclr/utils/flags.hpp
index f9505d31a7..16eecf48af 100644
--- a/rocclr/utils/flags.hpp
+++ b/rocclr/utils/flags.hpp
@@ -62,6 +62,8 @@ release(cstring, GPU_DEVICE_ORDINAL, "",                                      \
         "Select the device ordinal (comma seperated list of available devices)") \
 release(bool, REMOTE_ALLOC, false,                                            \
         "Use remote memory for the global heap allocation")                   \
+release(uint, GPU_CP_DMA_COPY_SIZE, 1,                                        \
+        "Set maximum size of CP DMA copy in KiB")                             \
 release(uint, GPU_MAX_HEAP_SIZE, 100,                                         \
         "Set maximum size of the GPU heap to % of board memory")              \
 release(uint, GPU_STAGING_BUFFER_SIZE, 1024,                                  \