diff --git a/projects/clr/rocclr/runtime/device/pal/palvirtual.cpp b/projects/clr/rocclr/runtime/device/pal/palvirtual.cpp
index a40d2f1e44..cd03ade47d 100644
--- a/projects/clr/rocclr/runtime/device/pal/palvirtual.cpp
+++ b/projects/clr/rocclr/runtime/device/pal/palvirtual.cpp
@@ -1909,9 +1909,13 @@ bool VirtualGPU::submitKernelInternal(const amd::NDRangeContainer& sizes, const
   }
 
   bool needFlush = false;
-  dmaFlushMgmt_.findSplitSize(dev(), sizes.global().product(), hsaKernel.aqlCodeSize());
-  if (dmaFlushMgmt().dispatchSplitSize() != 0) {
-    needFlush = true;
+
+  // Avoid flushing when PerfCounter is enabled, to make sure PerfStart/dispatch/PerfEnd are in the same cmdBuffer
+  if (!state_.perfCounterEnabled_) {
+    dmaFlushMgmt_.findSplitSize(dev(), sizes.global().product(), hsaKernel.aqlCodeSize());
+    if (dmaFlushMgmt().dispatchSplitSize() != 0) {
+      needFlush = true;
+    }
   }
 
   size_t newOffset[3] = {0, 0, 0};
@@ -2347,6 +2351,7 @@ void VirtualGPU::submitPerfCounter(amd::PerfCounterCommand& vcmd) {
       if (vcmd.getState() == amd::PerfCounterCommand::Begin) {
         Pal::SetClockModeInput input;
         Pal::SetClockModeOutput output = {};
+        state_.perfCounterEnabled_ = true;
         input.clockMode = Pal::DeviceClockMode::Profiling;
         dev().iDev()->SetClockMode(input, &output);
         GpuEvent event;
@@ -2360,6 +2365,7 @@ void VirtualGPU::submitPerfCounter(amd::PerfCounterCommand& vcmd) {
         iCmd()->CmdEndPerfExperiment(palPerf);
         eventEnd(MainEngine, event);
         setGpuEvent(event);
+        state_.perfCounterEnabled_ = false;
       } else {
         LogError("Unsupported performance counter state");
         vcmd.setStatus(CL_INVALID_OPERATION);
diff --git a/projects/clr/rocclr/runtime/device/pal/palvirtual.hpp b/projects/clr/rocclr/runtime/device/pal/palvirtual.hpp
index 65ee2f5ad1..4eabc1a902 100644
--- a/projects/clr/rocclr/runtime/device/pal/palvirtual.hpp
+++ b/projects/clr/rocclr/runtime/device/pal/palvirtual.hpp
@@ -180,9 +180,10 @@ class VirtualGPU : public device::VirtualDevice {
   //! The virtual GPU states
   union State {
     struct {
-      uint profiling_ : 1;       //!< Profiling is enabled
-      uint forceWait_ : 1;       //!< Forces wait in flush()
-      uint profileEnabled_ : 1;  //!< Profiling is enabled for WaveLimiter
+      uint profiling_          : 1;     //!< Profiling is enabled
+      uint forceWait_          : 1;     //!< Forces wait in flush()
+      uint profileEnabled_     : 1;     //!< Profiling is enabled for WaveLimiter
+      uint perfCounterEnabled_ : 1;     //!< PerfCounter is enabled
     };
     uint value_;
     State() : value_(0) {}