diff --git a/rocclr/runtime/device/gpu/gpuvirtual.cpp b/rocclr/runtime/device/gpu/gpuvirtual.cpp
index 3b0438c3c9..2754cf4af4 100644
--- a/rocclr/runtime/device/gpu/gpuvirtual.cpp
+++ b/rocclr/runtime/device/gpu/gpuvirtual.cpp
@@ -495,7 +495,8 @@ VirtualGPU::create(
             // hwRing_ should be set 0 if forced to have single scratch buffer
             hwRing_ = (dev().settings().useSingleScratch_) ? 0 : idx;
 
-            engineMask = dev().engines().getMask((gslEngineID)(GSL_ENGINEID_COMPUTE0 + idx));
+            engineMask = dev().engines().getMask((gslEngineID)(dev().isComputeRingIDForced() ?
+                         dev().getforcedComputeEngineID() : (GSL_ENGINEID_COMPUTE0 + idx)));
             if (dev().canDMA()) {
                 if (index() & 0x1) {
                     engineMask |= dev().engines().getMask(GSL_ENGINEID_DRMDMA0);
diff --git a/rocclr/runtime/device/gpu/gslbe/src/rt/GSLDevice.cpp b/rocclr/runtime/device/gpu/gslbe/src/rt/GSLDevice.cpp
index 5477bb2025..0f7c67a8b3 100644
--- a/rocclr/runtime/device/gpu/gslbe/src/rt/GSLDevice.cpp
+++ b/rocclr/runtime/device/gpu/gslbe/src/rt/GSLDevice.cpp
@@ -42,7 +42,8 @@ void CALGSLDevice::Initialize()
     m_vpuMask = 1;
     m_PerformLazyDeviceInit = false;
     m_computeRing = false;
-
+    m_isComputeRingIDForced = false;
+    m_forcedComputeEngineID = GSL_ENGINEID_INVALID;
     gslDeviceOps_ = NULL;
 }
 
@@ -313,8 +314,6 @@ CALGSLDevice::open(uint32 gpuIndex, bool enableHighPerformanceState, bool report
         return false;
     }
 
-    m_adp->queryAvailableEngines(&m_nEngines, m_engines);
-
     if (m_PerformLazyDeviceInit)
     {
         // close the adaptor
@@ -411,7 +410,9 @@ CALGSLDevice::SetupAdapter(int32 &asic_id)
 
     m_canDMA = hasDmaEngine;
 
-   //Disable DRMDMA on CFX mode for linux on all GPUs.
+    m_adp->queryAvailableEngines(&m_nEngines, m_engines);
+
+    //Disable DRMDMA on CFX mode for linux on all GPUs.
 #ifdef ATI_OS_LINUX
     if ((m_vpucount > 1) && !DRMDMA_FOR_LNX_CF)
     {
@@ -426,7 +427,7 @@ CALGSLDevice::SetupAdapter(int32 &asic_id)
         m_computeRing = false;
     }
     else
-    {   
+    {
         m_computeRing = true;
     }
 
@@ -435,6 +436,22 @@ CALGSLDevice::SetupAdapter(int32 &asic_id)
         m_computeRing = (GPU_NUM_COMPUTE_RINGS != 0);
     }
 
+    if ((!flagIsDefault(GPU_SELECT_COMPUTE_RINGS_ID)) && (m_computeRing))
+    {
+        gslEngineID engineID;
+        engineID = static_cast<gslEngineID>(GPU_SELECT_COMPUTE_RINGS_ID + GSL_ENGINEID_COMPUTE0);
+        if ((engineID >= GSL_ENGINEID_COMPUTE0) && (engineID <= GSL_ENGINEID_COMPUTE7))
+        {
+            for (uint i = 0; i < m_nEngines; ++i) {
+                if (m_engines[i].id == engineID){
+                    m_isComputeRingIDForced = true;
+                    m_forcedComputeEngineID = engineID;
+                    break;
+                }
+            }
+        }
+    }
+
     if (m_computeRing && !hasComputeEngine)
     {
         return false;
@@ -446,8 +463,8 @@ CALGSLDevice::SetupAdapter(int32 &asic_id)
 bool
 CALGSLDevice::SetupContext(int32 &asic_id)
 {
-    gsl::gsCtx* temp_cs = m_adp->createComputeContext(m_computeRing ? GSL_ENGINEID_COMPUTE0 : GSL_ENGINEID_3DCOMPUTE0, 
-                                                      m_canDMA ? GSL_ENGINEID_DRMDMA0 : GSL_ENGINEID_INVALID);
+    gsl::gsCtx* temp_cs = m_adp->createComputeContext(m_computeRing ? (m_isComputeRingIDForced ? m_forcedComputeEngineID : GSL_ENGINEID_COMPUTE0)
+                                                     : GSL_ENGINEID_3DCOMPUTE0, m_canDMA ? GSL_ENGINEID_DRMDMA0 : GSL_ENGINEID_INVALID);
     temp_cs->getMainSubCtx()->setVPUMask(m_vpuMask);
 
     m_revision = temp_cs->getChipRev();
@@ -632,8 +649,8 @@ CALGSLDevice::PerformFullInitialization_int()
 
     if (m_cs == 0)
     {
-        m_cs = m_adp->createComputeContext(m_computeRing ? GSL_ENGINEID_COMPUTE0 : GSL_ENGINEID_3DCOMPUTE0, 
-                                           m_canDMA ? GSL_ENGINEID_DRMDMA0 : GSL_ENGINEID_INVALID);
+        m_cs = m_adp->createComputeContext(m_computeRing ? (m_isComputeRingIDForced ? m_forcedComputeEngineID : GSL_ENGINEID_COMPUTE0)
+                                          : GSL_ENGINEID_3DCOMPUTE0, m_canDMA ? GSL_ENGINEID_DRMDMA0 : GSL_ENGINEID_INVALID);
         m_cs->getMainSubCtx()->setVPUMask(m_vpuMask);
 
         //
diff --git a/rocclr/runtime/device/gpu/gslbe/src/rt/GSLDevice.h b/rocclr/runtime/device/gpu/gslbe/src/rt/GSLDevice.h
index 9ab0f91ec8..bc0e87e17e 100644
--- a/rocclr/runtime/device/gpu/gslbe/src/rt/GSLDevice.h
+++ b/rocclr/runtime/device/gpu/gslbe/src/rt/GSLDevice.h
@@ -134,6 +134,9 @@ public:
 
     gsl::gsCtx* gslCtx() const { return m_cs; }
 
+    bool isComputeRingIDForced() const { return m_isComputeRingIDForced; }
+    gslEngineID getforcedComputeEngineID() const { return m_forcedComputeEngineID; }
+
 protected:
     //
     /// channel order enumerants
@@ -211,7 +214,7 @@ private:
     CALdevicestatus       m_deviceStatus;
     gslTextureResourceObject m_textureResource;
     gslSamplerObject      m_textureSampler;
-    bool                  m_isOpenCL200Device;
+    gslEngineID           m_forcedComputeEngineID;
 
     union {
         struct {
@@ -222,6 +225,8 @@ private:
             uint    m_PerformLazyDeviceInit : 1;
             uint    m_vmMode                : 1;
             uint    m_uavInCB               : 1;
+            uint    m_isComputeRingIDForced : 1;
+            bool    m_isOpenCL200Device     : 1;
         };
     };
 
diff --git a/rocclr/runtime/utils/flags.hpp b/rocclr/runtime/utils/flags.hpp
index 211b7dd323..b7b508c8cf 100644
--- a/rocclr/runtime/utils/flags.hpp
+++ b/rocclr/runtime/utils/flags.hpp
@@ -122,6 +122,8 @@ release(uint, GPU_SINGLE_ALLOC_PERCENT, 75,                                   \
         "Maximum size of a single allocation as percentage of total")         \
 release(uint, GPU_NUM_COMPUTE_RINGS, 2,                                       \
         "GPU number of compute rings. 0 - disabled, 1 , 2,.. - the number of compute rings") \
+release(int, GPU_SELECT_COMPUTE_RINGS_ID, -1,                                       \
+        "GPU select the compute rings ID -1 - disabled, 0 , 1,.. - the forced compute rings ID for submission") \
 release_on_stg(bool, C1X_ATOMICS, !IS_MAINLINE,                               \
         "Runtime will report c1x atomics support")                            \
 release(uint, GPU_WORKLOAD_SPLIT, 22,                                         \