diff --git a/projects/rocr-runtime/runtime/hsa-runtime/core/runtime/amd_gpu_agent.cpp b/projects/rocr-runtime/runtime/hsa-runtime/core/runtime/amd_gpu_agent.cpp
index b20b3edb5b..39b73292d6 100644
--- a/projects/rocr-runtime/runtime/hsa-runtime/core/runtime/amd_gpu_agent.cpp
+++ b/projects/rocr-runtime/runtime/hsa-runtime/core/runtime/amd_gpu_agent.cpp
@@ -717,7 +717,13 @@ core::Queue* GpuAgent::CreateInterceptibleQueue(void (*callback)(hsa_status_t st
 core::Blit* GpuAgent::CreateBlitSdma(bool use_xgmi, int rec_eng) {
   AMD::BlitSdmaBase* sdma;
   size_t copy_size_override = 0;
-  const size_t copy_size_overrides[2] = {0x3fffff, 0x3fffffff};
+  constexpr size_t copy_size_overrides[2] = {0x3fffff, 0x3fffffff};
+
+  /*
+   * On DXG SDMA packets placed in the queue are wrapped inside GCR packets by
+   * underlying driver and submitted into another queue. So GCR is not needed.
+   */
+  auto isDXG = core::Runtime::runtime_singleton_->thunkLoader()->IsDXG();
 
   switch (isa_->GetMajorVersion()) {
     case 9:
@@ -726,17 +732,13 @@ core::Blit* GpuAgent::CreateBlitSdma(bool use_xgmi, int rec_eng) {
                             copy_size_overrides[1] : copy_size_overrides[0];
       break;
     case 10:
-      sdma = new BlitSdmaV5();
+      sdma = (isDXG ? static_cast<BlitSdmaBase*>(new BlitSdmaV4()) : static_cast<BlitSdmaBase*>(new BlitSdmaV5()));
       copy_size_override = isa_->GetMinorVersion() < 3 ? copy_size_overrides[0] :
                                                          copy_size_overrides[1];
       break;
     case 11:
     case 12:
-      if (core::Runtime::runtime_singleton_->thunkLoader()->IsDXG()) {
-        sdma = new BlitSdmaV4();
-      } else {
-        sdma = new BlitSdmaV5();
-      }
+      sdma = (isDXG ? static_cast<BlitSdmaBase*>(new BlitSdmaV4()) : static_cast<BlitSdmaBase*>(new BlitSdmaV5()));
       copy_size_override = copy_size_overrides[1];
       break;
     default: