diff --git a/projects/clr/rocclr/device/rocm/rocvirtual.cpp b/projects/clr/rocclr/device/rocm/rocvirtual.cpp
index 8429355360..88a33cbdfe 100644
--- a/projects/clr/rocclr/device/rocm/rocvirtual.cpp
+++ b/projects/clr/rocclr/device/rocm/rocvirtual.cpp
@@ -127,6 +127,9 @@ void Timestamp::checkGpuTime() {
   if (HwProfiling()) {
     uint64_t start = std::numeric_limits<uint64_t>::max();
     uint64_t end = 0;
+    uint64_t sdmaStart = std::numeric_limits<uint64_t>::max();
+    uint64_t sdmaEnd = 0;
+
 
     for (auto it : signals_) {
       amd::ScopedLock lock(it->LockSignalOps());
@@ -140,41 +143,45 @@ void Timestamp::checkGpuTime() {
       if (command().GetBatchHead() == nullptr || command().profilingInfo().marker_ts_ ||
           command().type() == CL_COMMAND_TASK) {
         hsa_amd_profiling_dispatch_time_t time = {};
+        hsa_amd_profiling_async_copy_time_t timeSdma = {};
         amd_signal_t* amdSignal = reinterpret_cast<amd_signal_t*>(it->signal_.handle);
 
-        if (it->engine_ == HwQueueEngine::Compute) {
-          Hsa::profiling_get_dispatch_time(gpu()->gpu_device(), it->signal_, &time);
+        if (it->engine_ == HwQueueEngine::SdmaInter || it->engine_ == HwQueueEngine::SdmaRead ||
+            it->engine_ == HwQueueEngine::SdmaWrite || it->engine_ == HwQueueEngine::SdmaIntra) {
+          Hsa::profiling_get_async_copy_time(it->signal_, &timeSdma);
+          sdmaStart = std::min(timeSdma.start, sdmaStart);
+          sdmaEnd = std::max(timeSdma.end, sdmaEnd);
+          // set dispatch time to be used in logging.
+          time.start = timeSdma.start;
+          time.end = timeSdma.end;
         } else {
-          hsa_amd_profiling_async_copy_time_t time_sdma = {};
-          Hsa::profiling_get_async_copy_time(it->signal_, &time_sdma);
-          time.start = time_sdma.start;
-          time.end = time_sdma.end;
+          Hsa::profiling_get_dispatch_time(gpu()->gpu_device(), it->signal_, &time);
+          start = std::min(time.start, start);
+          end = std::max(time.end, end);
         }
 
-        start = std::min(time.start, start);
-        end = std::max(time.end, end);
-
         if ((command().type() == CL_COMMAND_TASK) && (it->flags_.isPacketDispatch_ == true)) {
           static_cast<amd::AccumulateCommand&>(command()).addTimestamps(time.start, time.end);
         }
 
         ClPrint(amd::LOG_INFO, amd::LOG_TS,
-                "Signal = (0x%lx), Translated start/end = %ld / %ld, "
-                "Elapsed = %ld ns, ticks start/end = %ld / %ld, Ticks elapsed = %ld",
+                "Signal = (0x%lx), Translated start/end = %ld / %ld, Elapsed = %ld ns, "
+                "ticks start/end = %ld / %ld, Ticks elapsed = %ld, Engine = %u",
                 it->signal_.handle, time.start, time.end, time.end - time.start,
-                amdSignal->start_ts, amdSignal->end_ts, amdSignal->end_ts - amdSignal->start_ts);
+                amdSignal->start_ts, amdSignal->end_ts, amdSignal->end_ts - amdSignal->start_ts,
+                it->engine_);
       }
       it->flags_.done_ = true;
     }
     signals_.clear();
-    if (end != 0) {
+    if (end != 0 || sdmaEnd != 0) {
       // Check if it's the first execution and update start time
       if (!accum_ena_) {
-        start_ = start * ticksToTime_;
+        start_ = ((sdmaEnd != 0) ? sdmaStart : start) * ticksToTime_;
         accum_ena_ = true;
       }
       // Progress the end time always
-      end_ = end * ticksToTime_;
+      end_ = ((sdmaEnd != 0) ? sdmaEnd : end) * ticksToTime_;
     }
   }
 }