diff --git a/runtime/hsa-runtime/CMakeLists.txt b/runtime/hsa-runtime/CMakeLists.txt
index 20d6d43450..a0f08ae9cd 100644
--- a/runtime/hsa-runtime/CMakeLists.txt
+++ b/runtime/hsa-runtime/CMakeLists.txt
@@ -87,7 +87,7 @@ if (ROCM_CCACHE_BUILD)
 endif() # if (ROCM_CCACHE_BUILD)
 
 ## Get version strings
-get_version ( "1.6.0" )
+get_version ( "1.7.0" )
 if ( ${ROCM_PATCH_VERSION} )
   set ( VERSION_PATCH ${ROCM_PATCH_VERSION})
 endif()
diff --git a/runtime/hsa-runtime/core/inc/memory_region.h b/runtime/hsa-runtime/core/inc/memory_region.h
index 4f362fad2d..30fbc07ac1 100644
--- a/runtime/hsa-runtime/core/inc/memory_region.h
+++ b/runtime/hsa-runtime/core/inc/memory_region.h
@@ -89,6 +89,7 @@ class MemoryRegion : public Checked<0x9C961F19EE175BB3> {
     AllocateDoubleMap = (1 << 2),   // Map twice VA allocation to backing store
     AllocateDirect = (1 << 3),      // Bypass fragment cache.
     AllocateIPC = (1 << 4),         // System memory that can be IPC-shared
+    AllocatePCIeRW = (1 << 5),      // Enforce pseudo fine grain/RW memory
   };
 
   typedef uint32_t AllocateFlags;
diff --git a/runtime/hsa-runtime/core/runtime/amd_memory_region.cpp b/runtime/hsa-runtime/core/runtime/amd_memory_region.cpp
index 7cbb543957..8327dec6cb 100644
--- a/runtime/hsa-runtime/core/runtime/amd_memory_region.cpp
+++ b/runtime/hsa-runtime/core/runtime/amd_memory_region.cpp
@@ -188,6 +188,9 @@ hsa_status_t MemoryRegion::AllocateImpl(size_t& size, AllocateFlags alloc_flags,
   if (IsSystem() && (alloc_flags & AllocateIPC))
       kmt_alloc_flags.ui32.NonPaged = 1;
 
+  // Allocate pseudo fine grain memory
+  kmt_alloc_flags.ui32.CoarseGrain = (alloc_flags & AllocatePCIeRW ? 0 : kmt_alloc_flags.ui32.CoarseGrain);
+
   // Only allow using the suballocator for ordinary VRAM.
   if (IsLocalMemory()) {
     bool subAllocEnabled = !core::Runtime::runtime_singleton_->flag().disable_fragment_alloc();
@@ -451,8 +454,7 @@ hsa_amd_memory_pool_access_t MemoryRegion::GetAccessInfo(
   // without regard to type of requesting device (CPU / GPU)
   // Return disallowed by default if framebuffer is fine grained
   // and requesting device is connected via xGMI link
-  // Return never allowed if framebuffer is fine grained and
-  // requesting device is connected via PCIe link
+
   if (IsLocalMemory()) {
 
     // Return disallowed by default if memory is coarse
@@ -461,12 +463,6 @@ hsa_amd_memory_pool_access_t MemoryRegion::GetAccessInfo(
       return HSA_AMD_MEMORY_POOL_ACCESS_DISALLOWED_BY_DEFAULT;
     }
 
-    // Determine if pool is pseudo fine-grained due to env flag
-    // Return disallowed by default
-    if (core::Runtime::runtime_singleton_->flag().fine_grain_pcie()) {
-      return HSA_AMD_MEMORY_POOL_ACCESS_DISALLOWED_BY_DEFAULT;
-    }
-
     // Return disallowed by default if memory is fine
     // grained and link type is xGMI.
     if (agent.HiveId() == owner()->HiveId()) {
diff --git a/runtime/hsa-runtime/core/runtime/hsa_ext_amd.cpp b/runtime/hsa-runtime/core/runtime/hsa_ext_amd.cpp
index 8ac975894d..2d39933577 100644
--- a/runtime/hsa-runtime/core/runtime/hsa_ext_amd.cpp
+++ b/runtime/hsa-runtime/core/runtime/hsa_ext_amd.cpp
@@ -686,7 +686,7 @@ hsa_status_t hsa_amd_memory_pool_allocate(hsa_amd_memory_pool_t memory_pool, siz
   TRY;
   IS_OPEN();
 
-  if (size == 0 || ptr == NULL || flags != 0) {
+  if (size == 0 || ptr == NULL || (flags > HSA_AMD_MEMORY_POOL_PCIE_FLAG)) {
     return HSA_STATUS_ERROR_INVALID_ARGUMENT;
   }
 
@@ -697,8 +697,11 @@ hsa_status_t hsa_amd_memory_pool_allocate(hsa_amd_memory_pool_t memory_pool, siz
     return (hsa_status_t)HSA_STATUS_ERROR_INVALID_MEMORY_POOL;
   }
 
-  return core::Runtime::runtime_singleton_->AllocateMemory(
-      mem_region, size, core::MemoryRegion::AllocateRestrict, ptr);
+  MemoryRegion::AllocateFlags alloc_flag = core::MemoryRegion::AllocateRestrict;
+
+  if (flags == HSA_AMD_MEMORY_POOL_PCIE_FLAG) alloc_flag |= core::MemoryRegion::AllocatePCIeRW;
+
+  return core::Runtime::runtime_singleton_->AllocateMemory(mem_region, size, alloc_flag, ptr);
   CATCH;
 }
 
diff --git a/runtime/hsa-runtime/inc/hsa_ext_amd.h b/runtime/hsa-runtime/inc/hsa_ext_amd.h
index 290c9ac845..9c9ef3d773 100644
--- a/runtime/hsa-runtime/inc/hsa_ext_amd.h
+++ b/runtime/hsa-runtime/inc/hsa_ext_amd.h
@@ -1000,6 +1000,24 @@ typedef enum {
   HSA_AMD_MEMORY_POOL_INFO_ALLOC_MAX_SIZE = 16,
 } hsa_amd_memory_pool_info_t;
 
+/**
+ * @brief Memory pool flag used to specify allocation directives
+ *
+ */
+typedef enum hsa_amd_memory_pool_flag_s {
+  /**
+   * Allocates memory that conforms to standard HSA memory consistency model
+   */
+  HSA_AMD_MEMORY_POOL_STANDARD_FLAG = 0,
+  /**
+   * Allocates fine grain memory type where memory ordering is per point to point
+   * connection. Atomic memory operations on these memory buffers are not
+   * guaranteed to be visible at system scope.
+   */
+  HSA_AMD_MEMORY_POOL_PCIE_FLAG = 1,
+
+} hsa_amd_memory_pool_flag_t;
+
 /**
  * @brief Get the current value of an attribute of a memory pool.
  *
@@ -1068,7 +1086,7 @@ hsa_status_t HSA_API hsa_amd_agent_iterate_memory_pools(
  * ::HSA_AMD_MEMORY_POOL_INFO_RUNTIME_ALLOC_GRANULE in @p memory_pool.
  *
  * @param[in] flags A bit-field that is used to specify allocation
- * directives. Reserved parameter, must be 0.
+ * directives.
  *
  * @param[out] ptr Pointer to the location where to store the base virtual
  * address of