diff --git a/projects/rocr-runtime/runtime/hsa-runtime/core/runtime/amd_topology.cpp b/projects/rocr-runtime/runtime/hsa-runtime/core/runtime/amd_topology.cpp
index 89c5cd2600..e595bffac7 100644
--- a/projects/rocr-runtime/runtime/hsa-runtime/core/runtime/amd_topology.cpp
+++ b/projects/rocr-runtime/runtime/hsa-runtime/core/runtime/amd_topology.cpp
@@ -367,6 +367,17 @@ void BuildTopology() {
   }
   const_cast<Flag&>(core::Runtime::runtime_singleton_->flag()).parse_masks(maxGpu, maxCu);
 
+  // Temporary work-around, disable SDMA ganging on non-APUs in non-SPX modes
+  // Check xGMI APU status
+  bool isXgmiApu = false;
+  auto& firstCpu = core::Runtime::runtime_singleton_->cpu_agents()[0];
+  for (auto& peer_gpu : core::Runtime::runtime_singleton_->gpu_agents()) {
+    auto linfo = core::Runtime::runtime_singleton_->GetLinkInfo(firstCpu->node_id(),
+                                                                peer_gpu->node_id());
+    isXgmiApu = linfo.info.link_type == HSA_AMD_LINK_INFO_TYPE_XGMI;
+    if (isXgmiApu) break;
+  }
+
   // Register destination agents that can SDMA gang copy for source agents
   for (auto& src_gpu : core::Runtime::runtime_singleton_->gpu_agents()) {
     uint32_t src_id = src_gpu->node_id();
@@ -383,7 +394,7 @@ void BuildTopology() {
         // Weight of 41 - Inter-socket GPU link in multi-partition mode
         if (linfo.info.link_type == HSA_AMD_LINK_INFO_TYPE_XGMI) {
           if (linfo.info.numa_distance == 13 || linfo.info.numa_distance == 41)
-            gang_factor = 2;
+            gang_factor = isXgmiApu ? 2 : 1;
           else if (linfo.info.numa_distance == 15 && linfo.info.min_bandwidth)
             gang_factor = linfo.info.max_bandwidth/linfo.info.min_bandwidth;
           else gang_factor = 1;