diff --git a/rocclr/device/blit.hpp b/rocclr/device/blit.hpp
index 165dc2359d..21b7b37c76 100644
--- a/rocclr/device/blit.hpp
+++ b/rocclr/device/blit.hpp
@@ -53,6 +53,7 @@ class BlitManager : public amd::HeapObject {
       uint disableFillImage_ : 1;
       uint disableCopyBufferToImageOpt_ : 1;
       uint disableHwlCopyBuffer_ : 1;
+      uint disableHostCopyBuffer_ : 1;
     };
     uint32_t value_;
     Setup() : value_(0) {}
diff --git a/rocclr/device/rocm/rocblit.cpp b/rocclr/device/rocm/rocblit.cpp
index 4be93dd4ad..1cdd232dac 100644
--- a/rocclr/device/rocm/rocblit.cpp
+++ b/rocclr/device/rocm/rocblit.cpp
@@ -69,7 +69,8 @@ bool DmaBlitManager::readBuffer(device::Memory& srcMemory, void* dstHost,
 
   // Use host copy if memory has direct access
   if (setup_.disableReadBuffer_ ||
-      (srcMemory.isHostMemDirectAccess() && !srcMemory.isCpuUncached())) {
+      (srcMemory.isHostMemDirectAccess() && !srcMemory.isCpuUncached() &&
+       !setup_.disableHostCopyBuffer_)) {
     // Stall GPU before CPU access
     gpu().Barriers().WaitCurrent();
     return HostBlitManager::readBuffer(srcMemory, dstHost, origin, size, entire, copyMetadata);
@@ -162,10 +163,12 @@ bool DmaBlitManager::readBufferRect(device::Memory& srcMemory, void* dstHost,
 
   // Use host copy if memory has direct access
   if (setup_.disableReadBufferRect_ ||
-      (srcMemory.isHostMemDirectAccess() && !srcMemory.isCpuUncached())) {
+      (srcMemory.isHostMemDirectAccess() && !srcMemory.isCpuUncached() &&
+       !setup_.disableHostCopyBuffer_)) {
     // Stall GPU before CPU access
     gpu().Barriers().WaitCurrent();
-    return HostBlitManager::readBufferRect(srcMemory, dstHost, bufRect, hostRect, size, entire, copyMetadata);
+    return HostBlitManager::readBufferRect(srcMemory, dstHost, bufRect, hostRect, size,
+                                           entire, copyMetadata);
   } else {
     Memory& xferBuf = dev().xferRead().acquire();
     address staging = xferBuf.getDeviceMemory();
@@ -233,8 +236,8 @@ bool DmaBlitManager::writeBuffer(const void* srcHost, device::Memory& dstMemory,
                                  const amd::Coord3D& origin, const amd::Coord3D& size,
                                  bool entire, amd::CopyMetadata copyMetadata) const {
   // Use host copy if memory has direct access
-  if (setup_.disableWriteBuffer_ || dstMemory.isHostMemDirectAccess() ||
-      gpuMem(dstMemory).IsPersistentDirectMap()) {
+  if ((setup_.disableWriteBuffer_ || dstMemory.isHostMemDirectAccess() ||
+      gpuMem(dstMemory).IsPersistentDirectMap()) && !setup_.disableHostCopyBuffer_) {
     // Stall GPU before CPU access
     gpu().releaseGpuMemoryFence();
     return HostBlitManager::writeBuffer(srcHost, dstMemory, origin, size, entire, copyMetadata);
@@ -330,8 +333,8 @@ bool DmaBlitManager::writeBufferRect(const void* srcHost, device::Memory& dstMem
   gpu().releaseGpuMemoryFence();
 
   // Use host copy if memory has direct access
-  if (setup_.disableWriteBufferRect_ || dstMemory.isHostMemDirectAccess() ||
-      gpuMem(dstMemory).IsPersistentDirectMap()) {
+  if ((setup_.disableWriteBufferRect_ || dstMemory.isHostMemDirectAccess() ||
+      gpuMem(dstMemory).IsPersistentDirectMap()) && !setup_.disableHostCopyBuffer_) {
     return HostBlitManager::writeBufferRect(srcHost, dstMemory, hostRect, bufRect, size, entire,
                                             copyMetadata);
   } else {
@@ -387,8 +390,9 @@ bool DmaBlitManager::copyBuffer(device::Memory& srcMemory, device::Memory& dstMe
                                 const amd::Coord3D& size, bool entire,
                                 amd::CopyMetadata copyMetadata) const {
   if (setup_.disableCopyBuffer_ ||
-      (srcMemory.isHostMemDirectAccess() && !srcMemory.isCpuUncached() &&
-      (dev().agent_profile() != HSA_PROFILE_FULL) && dstMemory.isHostMemDirectAccess())) {
+      (!setup_.disableHostCopyBuffer_ && srcMemory.isHostMemDirectAccess() &&
+      !srcMemory.isCpuUncached() && (dev().agent_profile() != HSA_PROFILE_FULL) &&
+      dstMemory.isHostMemDirectAccess())) {
     // Stall GPU before CPU access
     gpu().releaseGpuMemoryFence();
     return HostBlitManager::copyBuffer(srcMemory, dstMemory, srcOrigin, dstOrigin, size, false,
@@ -406,8 +410,8 @@ bool DmaBlitManager::copyBufferRect(device::Memory& srcMemory, device::Memory& d
                                     const amd::Coord3D& size, bool entire,
                                     amd::CopyMetadata copyMetadata) const {
   if (setup_.disableCopyBufferRect_ ||
-      (srcMemory.isHostMemDirectAccess() && !srcMemory.isCpuUncached() &&
-       dstMemory.isHostMemDirectAccess())) {
+      (!setup_.disableHostCopyBuffer_ && srcMemory.isHostMemDirectAccess() &&
+      !srcMemory.isCpuUncached() && dstMemory.isHostMemDirectAccess())) {
     // Stall GPU before CPU access
     gpu().releaseGpuMemoryFence();
     return HostBlitManager::copyBufferRect(srcMemory, dstMemory, srcRect, dstRect, size, entire,
@@ -1796,7 +1800,7 @@ bool KernelBlitManager::readBuffer(device::Memory& srcMemory, void* dstHost,
 
   // Use host copy if memory has direct access
   if (setup_.disableReadBuffer_ || (srcMemory.isHostMemDirectAccess() &&
-      !srcMemory.isCpuUncached())) {
+      !srcMemory.isCpuUncached() && !setup_.disableHostCopyBuffer_)) {
     // Stall GPU before CPU access
     gpu().releaseGpuMemoryFence();
     result = HostBlitManager::readBuffer(srcMemory, dstHost, origin, size, entire, copyMetadata);
@@ -1848,7 +1852,8 @@ bool KernelBlitManager::readBufferRect(device::Memory& srcMemory, void* dstHost,
 
   // Use host copy if memory has direct access
   if (setup_.disableReadBufferRect_ ||
-      (srcMemory.isHostMemDirectAccess() && !srcMemory.isCpuUncached())) {
+      (srcMemory.isHostMemDirectAccess() && !srcMemory.isCpuUncached() &&
+       !setup_.disableHostCopyBuffer_)) {
     // Stall GPU before CPU access
     gpu().releaseGpuMemoryFence();
     result = HostBlitManager::readBufferRect(srcMemory, dstHost, bufRect, hostRect, size, entire,
@@ -1899,7 +1904,7 @@ bool KernelBlitManager::writeBuffer(const void* srcHost, device::Memory& dstMemo
 
   // Use host copy if memory has direct access
   if (setup_.disableWriteBuffer_ || dstMemory.isHostMemDirectAccess() ||
-      gpuMem(dstMemory).IsPersistentDirectMap()) {
+      gpuMem(dstMemory).IsPersistentDirectMap() && !setup_.disableHostCopyBuffer_) {
     // Stall GPU before CPU access
     gpu().releaseGpuMemoryFence();
     result = HostBlitManager::writeBuffer(srcHost, dstMemory, origin, size, entire, copyMetadata);
@@ -1950,8 +1955,8 @@ bool KernelBlitManager::writeBufferRect(const void* srcHost, device::Memory& dst
   bool result = false;
 
   // Use host copy if memory has direct access
-  if (setup_.disableWriteBufferRect_ || dstMemory.isHostMemDirectAccess() ||
-      gpuMem(dstMemory).IsPersistentDirectMap()) {
+  if ((setup_.disableWriteBufferRect_ || dstMemory.isHostMemDirectAccess() ||
+      gpuMem(dstMemory).IsPersistentDirectMap()) && !setup_.disableHostCopyBuffer_) {
     // Stall GPU before CPU access
     gpu().releaseGpuMemoryFence();
     result = HostBlitManager::writeBufferRect(srcHost, dstMemory, hostRect, bufRect, size, entire,
@@ -2031,7 +2036,8 @@ bool KernelBlitManager::fillBuffer1D(device::Memory& memory, const void* pattern
   bool result = false;
 
   // Use host fill if memory has direct access
-  if (setup_.disableFillBuffer_ || (!forceBlit && memory.isHostMemDirectAccess())) {
+  if (setup_.disableFillBuffer_ || (!forceBlit && memory.isHostMemDirectAccess() &&
+      !setup_.disableHostCopyBuffer_)) {
     // Stall GPU before CPU access
     gpu().releaseGpuMemoryFence();
     result = HostBlitManager::fillBuffer(memory, pattern, patternSize, size, origin, size, entire);
@@ -2130,7 +2136,8 @@ bool KernelBlitManager::fillBuffer2D(device::Memory& memory, const void* pattern
   bool result = false;
 
     // Use host fill if memory has direct access
-  if (setup_.disableFillBuffer_ || (!forceBlit && memory.isHostMemDirectAccess())) {
+  if (setup_.disableFillBuffer_ || (!forceBlit && memory.isHostMemDirectAccess() &&
+      !setup_.disableHostCopyBuffer_)) {
     // Stall GPU before CPU access
     gpu().releaseGpuMemoryFence();
     result = HostBlitManager::fillBuffer(memory, pattern, patternSize, size, origin, size, entire);
diff --git a/rocclr/device/rocm/rocvirtual.cpp b/rocclr/device/rocm/rocvirtual.cpp
index dbac1d3df3..97c38d2437 100644
--- a/rocclr/device/rocm/rocvirtual.cpp
+++ b/rocclr/device/rocm/rocvirtual.cpp
@@ -1358,6 +1358,8 @@ bool VirtualGPU::create() {
   }
 
   device::BlitManager::Setup blitSetup;
+  // Disable HostBlit copy path for HIP
+  blitSetup.disableHostCopyBuffer_ = amd::IS_HIP;
   blitMgr_ = new KernelBlitManager(*this, blitSetup);
   if ((nullptr == blitMgr_) || !blitMgr_->create(roc_device_)) {
     LogError("Could not create BlitManager!");