diff --git a/projects/clr/rocclr/device/pal/palvirtual.cpp b/projects/clr/rocclr/device/pal/palvirtual.cpp
index 44b06436d6..0ae8b85afa 100644
--- a/projects/clr/rocclr/device/pal/palvirtual.cpp
+++ b/projects/clr/rocclr/device/pal/palvirtual.cpp
@@ -2487,7 +2487,7 @@ void VirtualGPU::submitKernel(amd::NDRangeKernelCommand& vcmd) {
 
     // Submit kernel to HW
     if (!queue->submitKernelInternal(vcmd.sizes(), vcmd.kernel(), vcmd.parameters(), false,
-                                     vcmd.sharedMemBytes(), vcmd.cooperativeGroups())) {
+                                     vcmd.sharedMemBytes())) {
       vcmd.setStatus(CL_INVALID_OPERATION);
     }
 
@@ -2503,7 +2503,7 @@ void VirtualGPU::submitKernel(amd::NDRangeKernelCommand& vcmd) {
 
     // Submit kernel to HW
     if (!submitKernelInternal(vcmd.sizes(), vcmd.kernel(), vcmd.parameters(), false,
-                              vcmd.sharedMemBytes(), vcmd.cooperativeGroups())) {
+                              vcmd.sharedMemBytes(), vcmd.getAnyOrderLaunchFlag())) {
       vcmd.setStatus(CL_INVALID_OPERATION);
     }
 
@@ -2515,9 +2515,10 @@ void VirtualGPU::submitKernel(amd::NDRangeKernelCommand& vcmd) {
 bool VirtualGPU::submitKernelInternal(const amd::NDRangeContainer& sizes,
                                       const amd::Kernel& kernel, const_address parameters,
                                       bool nativeMem, uint32_t sharedMemBytes,
-                                      bool cooperativeGroup) {
+                                      bool anyOrder) {
   size_t newOffset[3] = {0, 0, 0};
   size_t newGlobalSize[3] = {0, 0, 0};
+  state_.anyOrder_ = anyOrder;
 
   int dim = -1;
   int iteration = 1;
diff --git a/projects/clr/rocclr/device/pal/palvirtual.hpp b/projects/clr/rocclr/device/pal/palvirtual.hpp
index 8086e5eb2c..61f472ed91 100644
--- a/projects/clr/rocclr/device/pal/palvirtual.hpp
+++ b/projects/clr/rocclr/device/pal/palvirtual.hpp
@@ -240,6 +240,7 @@ class VirtualGPU : public device::VirtualDevice {
       uint perfCounterEnabled_ : 1;  //!< PerfCounter is enabled
       uint rgpCaptureEnabled_ : 1;   //!< RGP capture is enabled in the runtime
       uint imageBufferWrtBack_ : 1;  //!< Enable image buffer write back
+      uint anyOrder_ : 1;            //!< Kernel launches don't need a barrier
     };
     uint value_;
     State() : value_(0) {}
@@ -267,7 +268,12 @@ class VirtualGPU : public device::VirtualDevice {
     //! Invalidates GPU caches if memory dependency tracking is disabled
     void sync(VirtualGPU& gpu) const {
       if (maxMemObjectsInQueue_ == 0) {
-        gpu.addBarrier(RgpSqqtBarrierReason::MemDependency);
+        // Ignore the barrier in any order mode. The app is responsible for synchronization.
+        // HW will execute the kernels asynchronously
+        if (!gpu.anyOrder()) {
+          // Wait for GPU and invalidate L1 cache
+          gpu.addBarrier(RgpSqqtBarrierReason::MemDependency);
+        }
       }
     }
 
@@ -309,7 +315,7 @@ class VirtualGPU : public device::VirtualDevice {
       const_address parameters,            //!< Parameters for the kernel
       bool nativeMem = true,               //!< Native memory objects
       uint32_t sharedMemBytes = 0,         //!< Shared memory size
-      bool cooperativeGroups = false       //!< TRUE if cooperative groups mode is required
+      bool anyOrder = false                //!< TRUE if any order launch mode is enabled
   );
   void submitNativeFn(amd::NativeFnCommand& vcmd);
   void submitFillMemory(amd::FillMemoryCommand& vcmd);
@@ -434,6 +440,9 @@ class VirtualGPU : public device::VirtualDevice {
   //! Checks if profiling is enabled
   bool profiling() const { return state_.profiling_; }
 
+  //! Checks if the queue is in any order mode
+  bool anyOrder() const { return state_.anyOrder_; }
+
   //! Returns memory dependency class
   MemoryDependency& memoryDependency() { return memoryDependency_; }