diff --git a/subprojects/gst-plugins-bad/docs/libs/cuda/index.md b/subprojects/gst-plugins-bad/docs/libs/cuda/index.md
index 5e529291b8..abfe9b061e 100644
--- a/subprojects/gst-plugins-bad/docs/libs/cuda/index.md
+++ b/subprojects/gst-plugins-bad/docs/libs/cuda/index.md
@@ -23,3 +23,12 @@ Example: `GST_CUDA_CRITICAL_ERRORS=2,700`
 As a result of the above example, if `CUDA_ERROR_OUT_OF_MEMORY(2)` or
 `CUDA_ERROR_ILLEGAL_ADDRESS(700)` error is detected in GStreamer CUDA library,
 the process will be aborted.
+
+
+**`GST_CUDA_ENABLE_STREAM_ORDERED_ALLOC`. (Since: 1.26)**
+
+As of 1.26, GStreamer CUDA library supports stream ordered CUDA allocation
+(e.g., cuMemAllocAsync). The new allocation method is disabled by default
+unless it's explicitly requested via buffer pool option.
+This environment variable can be used to change the default behavior
+so that the stream ordered allocation can be used by default.
diff --git a/subprojects/gst-plugins-bad/gst-libs/gst/cuda/gstcudabufferpool.cpp b/subprojects/gst-plugins-bad/gst-libs/gst/cuda/gstcudabufferpool.cpp
index a5efd1fedc..49ea29ae80 100644
--- a/subprojects/gst-plugins-bad/gst-libs/gst/cuda/gstcudabufferpool.cpp
+++ b/subprojects/gst-plugins-bad/gst-libs/gst/cuda/gstcudabufferpool.cpp
@@ -86,6 +86,19 @@ gst_cuda_buffer_pool_update_alloc_prop (GstCudaBufferPool * self)
   return TRUE;
 }
 
+static gboolean
+default_stream_ordered_alloc_enabled (void)
+{
+  static gboolean enabled = FALSE;
+  GST_CUDA_CALL_ONCE_BEGIN {
+    if (g_getenv ("GST_CUDA_ENABLE_STREAM_ORDERED_ALLOC"))
+      enabled = TRUE;
+  }
+  GST_CUDA_CALL_ONCE_END;
+
+  return enabled;
+}
+
 static gboolean
 gst_cuda_buffer_pool_set_config (GstBufferPool * pool, GstStructure * config)
 {
@@ -133,8 +146,25 @@ gst_cuda_buffer_pool_set_config (GstBufferPool * pool, GstStructure * config)
     priv->alloc = gst_cuda_pool_allocator_new_for_virtual_memory (self->context,
         priv->stream, &info, &priv->prop, CU_MEM_ALLOC_GRANULARITY_MINIMUM);
   } else {
-    gboolean stream_ordered =
-        gst_buffer_pool_config_get_cuda_stream_ordered_alloc (config);
+    gboolean stream_ordered = FALSE;
+    if (!gst_buffer_pool_config_get_cuda_stream_ordered_alloc (config,
+            &stream_ordered)) {
+      gboolean prefer_stream_ordered = FALSE;
+      g_object_get (self->context, "prefer-stream-ordered-alloc",
+          &prefer_stream_ordered, nullptr);
+      if (prefer_stream_ordered) {
+        GST_DEBUG_OBJECT (self, "Stream ordered alloc was enabled in context");
+        stream_ordered = TRUE;
+      } else {
+        stream_ordered = default_stream_ordered_alloc_enabled ();
+        GST_DEBUG_OBJECT (self, "Use default stream ordered alloc: %d",
+            stream_ordered);
+      }
+    } else {
+      GST_DEBUG_OBJECT (self,
+          "stream ordered alloc by config: %d", stream_ordered);
+    }
+
     GstStructure *alloc_config = gst_structure_new ("alloc-config",
         GST_CUDA_ALLOCATOR_OPT_STREAM_ORDERED, G_TYPE_BOOLEAN, stream_ordered,
         nullptr);
@@ -343,13 +373,15 @@ gst_buffer_pool_config_set_cuda_alloc_method (GstStructure * config,
 /**
  * gst_buffer_pool_config_get_cuda_stream_ordered_alloc:
  * @config: a buffer pool config
+ * @enabled: (out): whether stream ordered allocation was requested or not
  *
- * Returns: %TRUE if stream ordered allocation is enabled
+ * Returns: %TRUE stream ordered allocation option was specified
  *
  * Since: 1.26
  */
 gboolean
-gst_buffer_pool_config_get_cuda_stream_ordered_alloc (GstStructure * config)
+gst_buffer_pool_config_get_cuda_stream_ordered_alloc (GstStructure * config,
+    gboolean * enabled)
 {
   gboolean stream_ordered = FALSE;
 
@@ -357,11 +389,13 @@ gst_buffer_pool_config_get_cuda_stream_ordered_alloc (GstStructure * config)
 
   if (!gst_structure_get_boolean (config,
           "cuda-stream-ordered-alloc", &stream_ordered)) {
-    /* Disable stream-ordered alloc by default */
     return FALSE;
   }
 
-  return stream_ordered;
+  if (enabled)
+    *enabled = stream_ordered;
+
+  return TRUE;
 }
 
 /**
diff --git a/subprojects/gst-plugins-bad/gst-libs/gst/cuda/gstcudabufferpool.h b/subprojects/gst-plugins-bad/gst-libs/gst/cuda/gstcudabufferpool.h
index 2c42130338..8dc8f54499 100644
--- a/subprojects/gst-plugins-bad/gst-libs/gst/cuda/gstcudabufferpool.h
+++ b/subprojects/gst-plugins-bad/gst-libs/gst/cuda/gstcudabufferpool.h
@@ -87,7 +87,8 @@ void                     gst_buffer_pool_config_set_cuda_alloc_method (GstStruct
                                                                        GstCudaMemoryAllocMethod method);
 
 GST_CUDA_API
-gboolean        gst_buffer_pool_config_get_cuda_stream_ordered_alloc (GstStructure * config);
+gboolean        gst_buffer_pool_config_get_cuda_stream_ordered_alloc (GstStructure * config,
+                                                                      gboolean * enabled);
 
 GST_CUDA_API
 void            gst_buffer_pool_config_set_cuda_stream_ordered_alloc (GstStructure * config,
diff --git a/subprojects/gst-plugins-bad/gst-libs/gst/cuda/gstcudacontext.cpp b/subprojects/gst-plugins-bad/gst-libs/gst/cuda/gstcudacontext.cpp
index ab6e0afa09..e5cc64925e 100644
--- a/subprojects/gst-plugins-bad/gst-libs/gst/cuda/gstcudacontext.cpp
+++ b/subprojects/gst-plugins-bad/gst-libs/gst/cuda/gstcudacontext.cpp
@@ -55,6 +55,7 @@ enum
   PROP_VIRTUAL_MEMORY,
   PROP_OS_HANDLE,
   PROP_STREAM_ORDERED_ALLOC,
+  PROP_PREFER_STREAM_ORDERED_ALLLOC,
 };
 
 struct _GstCudaContextPrivate
@@ -66,11 +67,14 @@ struct _GstCudaContextPrivate
   gboolean virtual_memory_supported;
   gboolean os_handle_supported;
   gboolean stream_ordered_alloc_supported;
+  gboolean prefer_stream_ordered_alloc;
 
   gint tex_align;
 
   GHashTable *accessible_peer;
   gboolean owns_context;
+
+  GMutex lock;
 };
 
 #define gst_cuda_context_parent_class parent_class
@@ -152,6 +156,17 @@ gst_cuda_context_class_init (GstCudaContextClass * klass)
           "Device supports stream ordered allocation", FALSE,
           (GParamFlags) (G_PARAM_READABLE | G_PARAM_STATIC_STRINGS)));
 
+  /**
+   * GstCudaContext:prefer-stream-ordered-alloc:
+   *
+   * Since: 1.26
+   */
+  g_object_class_install_property (gobject_class,
+      PROP_PREFER_STREAM_ORDERED_ALLLOC,
+      g_param_spec_boolean ("prefer-stream-ordered-alloc",
+          "Prefer Stream Ordered Alloc", "Prefers stream ordered allocation",
+          FALSE, (GParamFlags) (G_PARAM_READWRITE | G_PARAM_STATIC_STRINGS)));
+
   gst_cuda_memory_init_once ();
 }
 
@@ -162,6 +177,7 @@ gst_cuda_context_init (GstCudaContext * context)
       gst_cuda_context_get_instance_private (context);
 
   priv->accessible_peer = g_hash_table_new (g_direct_hash, g_direct_equal);
+  g_mutex_init (&priv->lock);
 
   context->priv = priv;
 }
@@ -173,10 +189,16 @@ gst_cuda_context_set_property (GObject * object, guint prop_id,
   GstCudaContext *context = GST_CUDA_CONTEXT (object);
   GstCudaContextPrivate *priv = context->priv;
 
+
   switch (prop_id) {
     case PROP_DEVICE_ID:
       priv->device_id = g_value_get_uint (value);
       break;
+    case PROP_PREFER_STREAM_ORDERED_ALLLOC:
+      g_mutex_lock (&priv->lock);
+      priv->prefer_stream_ordered_alloc = g_value_get_boolean (value);
+      g_mutex_unlock (&priv->lock);
+      break;
     default:
       G_OBJECT_WARN_INVALID_PROPERTY_ID (object, prop_id, pspec);
       break;
@@ -206,6 +228,11 @@ gst_cuda_context_get_property (GObject * object, guint prop_id,
     case PROP_STREAM_ORDERED_ALLOC:
       g_value_set_boolean (value, priv->stream_ordered_alloc_supported);
       break;
+    case PROP_PREFER_STREAM_ORDERED_ALLLOC:
+      g_mutex_lock (&priv->lock);
+      g_value_set_boolean (value, priv->prefer_stream_ordered_alloc);
+      g_mutex_unlock (&priv->lock);
+      break;
     default:
       G_OBJECT_WARN_INVALID_PROPERTY_ID (object, prop_id, pspec);
       break;
@@ -419,6 +446,8 @@ gst_cuda_context_finalize (GObject * object)
     gst_cuda_result (CuCtxDestroy (priv->context));
   }
 
+  g_mutex_clear (&priv->lock);
+
   G_OBJECT_CLASS (parent_class)->finalize (object);
 }