diff --git a/subprojects/gst-plugins-bad/docs/plugins/gst_plugins_cache.json b/subprojects/gst-plugins-bad/docs/plugins/gst_plugins_cache.json
index 9cbfc35d04..5dd045db2b 100644
--- a/subprojects/gst-plugins-bad/docs/plugins/gst_plugins_cache.json
+++ b/subprojects/gst-plugins-bad/docs/plugins/gst_plugins_cache.json
@@ -225402,12 +225402,12 @@
                         "writable": true
                     },
                     "num-output-surfaces": {
-                        "blurb": "Maximum number of output surfaces simultaneously mapped in CUDA output mode (0 = auto)",
+                        "blurb": "Maximum number of output surfaces simultaneously mapped in CUDA output mode (0 = auto, 1 = always copy)",
                         "conditionally-available": false,
                         "construct": false,
                         "construct-only": false,
                         "controllable": false,
-                        "default": "0",
+                        "default": "1",
                         "max": "64",
                         "min": "0",
                         "mutable": "ready",
@@ -227333,12 +227333,12 @@
                         "writable": true
                     },
                     "num-output-surfaces": {
-                        "blurb": "Maximum number of output surfaces simultaneously mapped in CUDA output mode (0 = auto)",
+                        "blurb": "Maximum number of output surfaces simultaneously mapped in CUDA output mode (0 = auto, 1 = always copy)",
                         "conditionally-available": false,
                         "construct": false,
                         "construct-only": false,
                         "controllable": false,
-                        "default": "0",
+                        "default": "1",
                         "max": "64",
                         "min": "0",
                         "mutable": "ready",
@@ -227555,12 +227555,12 @@
                         "writable": true
                     },
                     "num-output-surfaces": {
-                        "blurb": "Maximum number of output surfaces simultaneously mapped in CUDA output mode (0 = auto)",
+                        "blurb": "Maximum number of output surfaces simultaneously mapped in CUDA output mode (0 = auto, 1 = always copy)",
                         "conditionally-available": false,
                         "construct": false,
                         "construct-only": false,
                         "controllable": false,
-                        "default": "0",
+                        "default": "1",
                         "max": "64",
                         "min": "0",
                         "mutable": "ready",
@@ -227944,12 +227944,12 @@
                         "writable": true
                     },
                     "num-output-surfaces": {
-                        "blurb": "Maximum number of output surfaces simultaneously mapped in CUDA output mode (0 = auto)",
+                        "blurb": "Maximum number of output surfaces simultaneously mapped in CUDA output mode (0 = auto, 1 = always copy)",
                         "conditionally-available": false,
                         "construct": false,
                         "construct-only": false,
                         "controllable": false,
-                        "default": "0",
+                        "default": "1",
                         "max": "64",
                         "min": "0",
                         "mutable": "ready",
@@ -228043,12 +228043,12 @@
                         "writable": true
                     },
                     "num-output-surfaces": {
-                        "blurb": "Maximum number of output surfaces simultaneously mapped in CUDA output mode (0 = auto)",
+                        "blurb": "Maximum number of output surfaces simultaneously mapped in CUDA output mode (0 = auto, 1 = always copy)",
                         "conditionally-available": false,
                         "construct": false,
                         "construct-only": false,
                         "controllable": false,
-                        "default": "0",
+                        "default": "1",
                         "max": "64",
                         "min": "0",
                         "mutable": "ready",
diff --git a/subprojects/gst-plugins-bad/sys/nvcodec/gstnvav1dec.cpp b/subprojects/gst-plugins-bad/sys/nvcodec/gstnvav1dec.cpp
index c943ff8a3d..b9a8cffda6 100644
--- a/subprojects/gst-plugins-bad/sys/nvcodec/gstnvav1dec.cpp
+++ b/subprojects/gst-plugins-bad/sys/nvcodec/gstnvav1dec.cpp
@@ -94,7 +94,7 @@ enum
   PROP_MAX_DISPLAY_DELAY,
 };
 
-#define DEFAULT_NUM_OUTPUT_SURFACES 0
+#define DEFAULT_NUM_OUTPUT_SURFACES 1
 #define DEFAULT_MAX_DISPLAY_DELAY -1
 
 static GTypeClass *parent_class = nullptr;
@@ -162,16 +162,17 @@ gst_nv_av1_dec_class_init (GstNvAV1DecClass * klass,
   /**
    * GstNvAV1Dec:num-output-surfaces:
    *
-   * The number of output surfaces (0 = auto). This property will be used to
-   * calculate the CUVIDDECODECREATEINFO.ulNumOutputSurfaces parameter
-   * in case of CUDA output mode
+   * The number of output surfaces (0 = auto, 1 = always copy).
+   * This property will be used to calculate
+   * the CUVIDDECODECREATEINFO.ulNumOutputSurfaces parameter in case of
+   * CUDA output mode.
    *
    * Since: 1.24
    */
   g_object_class_install_property (object_class, PROP_NUM_OUTPUT_SURFACES,
       g_param_spec_uint ("num-output-surfaces", "Num Output Surfaces",
           "Maximum number of output surfaces simultaneously mapped in CUDA "
-          "output mode (0 = auto)",
+          "output mode (0 = auto, 1 = always copy)",
           0, 64, DEFAULT_NUM_OUTPUT_SURFACES,
           (GParamFlags) (GST_PARAM_MUTABLE_READY | G_PARAM_READWRITE |
               G_PARAM_STATIC_STRINGS)));
diff --git a/subprojects/gst-plugins-bad/sys/nvcodec/gstnvdecoder.cpp b/subprojects/gst-plugins-bad/sys/nvcodec/gstnvdecoder.cpp
index 9ffe660748..3f7b45e00f 100644
--- a/subprojects/gst-plugins-bad/sys/nvcodec/gstnvdecoder.cpp
+++ b/subprojects/gst-plugins-bad/sys/nvcodec/gstnvdecoder.cpp
@@ -411,9 +411,8 @@ gst_nv_decoder_new_picture (GstNvDecoder * decoder, GstCodecPicture * picture)
 
   if (!decoder->object) {
     if (decoder->output_type == GST_NV_DECODER_OUTPUT_TYPE_CUDA) {
-      if (decoder->num_output_surfaces == 0 ||
-          decoder->num_output_surfaces < decoder->downstream_min_buffers) {
-        /* Auto mode or user specified num-output-surfaces value is too small */
+      if (decoder->num_output_surfaces == 0) {
+        /* Auto mode */
         decoder->create_info.ulNumOutputSurfaces =
             decoder->downstream_min_buffers + 2;
       } else {
@@ -877,7 +876,7 @@ gst_nv_decoder_output_picture (GstNvDecoder * decoder,
     goto error;
   }
 
-  if (videodec->input_segment.rate > 0 &&
+  if (decoder->num_output_surfaces != 1 && videodec->input_segment.rate > 0 &&
       decoder->output_type == GST_NV_DECODER_OUTPUT_TYPE_CUDA &&
       (guint) decoder->create_info.ulNumOutputSurfaces >=
       decoder->downstream_min_buffers) {
diff --git a/subprojects/gst-plugins-bad/sys/nvcodec/gstnvh264dec.cpp b/subprojects/gst-plugins-bad/sys/nvcodec/gstnvh264dec.cpp
index 0d39163759..66a4e2355a 100644
--- a/subprojects/gst-plugins-bad/sys/nvcodec/gstnvh264dec.cpp
+++ b/subprojects/gst-plugins-bad/sys/nvcodec/gstnvh264dec.cpp
@@ -150,7 +150,7 @@ enum
   PROP_MAX_DISPLAY_DELAY,
 };
 
-#define DEFAULT_NUM_OUTPUT_SURFACES 0
+#define DEFAULT_NUM_OUTPUT_SURFACES 1
 #define DEFAULT_MAX_DISPLAY_DELAY -1
 
 static GTypeClass *parent_class = nullptr;
@@ -229,16 +229,17 @@ gst_nv_h264_dec_class_init (GstNvH264DecClass * klass,
   /**
    * GstNvH264Dec:num-output-surfaces:
    *
-   * The number of output surfaces (0 = auto). This property will be used to
-   * calculate the CUVIDDECODECREATEINFO.ulNumOutputSurfaces parameter
-   * in case of CUDA output mode
+   * The number of output surfaces (0 = auto, 1 = always copy).
+   * This property will be used to calculate
+   * the CUVIDDECODECREATEINFO.ulNumOutputSurfaces parameter in case of
+   * CUDA output mode.
    *
    * Since: 1.24
    */
   g_object_class_install_property (object_class, PROP_NUM_OUTPUT_SURFACES,
       g_param_spec_uint ("num-output-surfaces", "Num Output Surfaces",
           "Maximum number of output surfaces simultaneously mapped in CUDA "
-          "output mode (0 = auto)",
+          "output mode (0 = auto, 1 = always copy)",
           0, 64, DEFAULT_NUM_OUTPUT_SURFACES,
           (GParamFlags) (GST_PARAM_MUTABLE_READY | G_PARAM_READWRITE |
               G_PARAM_STATIC_STRINGS)));
diff --git a/subprojects/gst-plugins-bad/sys/nvcodec/gstnvh265dec.cpp b/subprojects/gst-plugins-bad/sys/nvcodec/gstnvh265dec.cpp
index 313e956311..5d8e46404d 100644
--- a/subprojects/gst-plugins-bad/sys/nvcodec/gstnvh265dec.cpp
+++ b/subprojects/gst-plugins-bad/sys/nvcodec/gstnvh265dec.cpp
@@ -149,7 +149,7 @@ enum
   PROP_MAX_DISPLAY_DELAY,
 };
 
-#define DEFAULT_NUM_OUTPUT_SURFACES 0
+#define DEFAULT_NUM_OUTPUT_SURFACES 1
 #define DEFAULT_MAX_DISPLAY_DELAY -1
 
 static GTypeClass *parent_class = nullptr;
@@ -225,16 +225,17 @@ gst_nv_h265_dec_class_init (GstNvH265DecClass * klass,
   /**
    * GstNvH265Dec:num-output-surfaces:
    *
-   * The number of output surfaces (0 = auto). This property will be used to
-   * calculate the CUVIDDECODECREATEINFO.ulNumOutputSurfaces parameter
-   * in case of CUDA output mode
+   * The number of output surfaces (0 = auto, 1 = always copy).
+   * This property will be used to calculate
+   * the CUVIDDECODECREATEINFO.ulNumOutputSurfaces parameter in case of
+   * CUDA output mode.
    *
    * Since: 1.24
    */
   g_object_class_install_property (object_class, PROP_NUM_OUTPUT_SURFACES,
       g_param_spec_uint ("num-output-surfaces", "Num Output Surfaces",
           "Maximum number of output surfaces simultaneously mapped in CUDA "
-          "output mode (0 = auto)",
+          "output mode (0 = auto, 1 = always copy)",
           0, 64, DEFAULT_NUM_OUTPUT_SURFACES,
           (GParamFlags) (GST_PARAM_MUTABLE_READY | G_PARAM_READWRITE |
               G_PARAM_STATIC_STRINGS)));
diff --git a/subprojects/gst-plugins-bad/sys/nvcodec/gstnvvp8dec.cpp b/subprojects/gst-plugins-bad/sys/nvcodec/gstnvvp8dec.cpp
index 23cc792eb1..0273590ee1 100644
--- a/subprojects/gst-plugins-bad/sys/nvcodec/gstnvvp8dec.cpp
+++ b/subprojects/gst-plugins-bad/sys/nvcodec/gstnvvp8dec.cpp
@@ -78,7 +78,7 @@ enum
   PROP_MAX_DISPLAY_DELAY,
 };
 
-#define DEFAULT_NUM_OUTPUT_SURFACES 0
+#define DEFAULT_NUM_OUTPUT_SURFACES 1
 #define DEFAULT_MAX_DISPLAY_DELAY -1
 
 static GTypeClass *parent_class = nullptr;
@@ -148,16 +148,17 @@ gst_nv_vp8_dec_class_init (GstNvVp8DecClass * klass,
   /**
    * GstNvVp8Dec:num-output-surfaces:
    *
-   * The number of output surfaces (0 = auto). This property will be used to
-   * calculate the CUVIDDECODECREATEINFO.ulNumOutputSurfaces parameter
-   * in case of CUDA output mode
+   * The number of output surfaces (0 = auto, 1 = always copy).
+   * This property will be used to calculate
+   * the CUVIDDECODECREATEINFO.ulNumOutputSurfaces parameter in case of
+   * CUDA output mode.
    *
    * Since: 1.24
    */
   g_object_class_install_property (object_class, PROP_NUM_OUTPUT_SURFACES,
       g_param_spec_uint ("num-output-surfaces", "Num Output Surfaces",
           "Maximum number of output surfaces simultaneously mapped in CUDA "
-          "output mode (0 = auto)",
+          "output mode (0 = auto, 1 = always copy)",
           0, 64, DEFAULT_NUM_OUTPUT_SURFACES,
           (GParamFlags) (GST_PARAM_MUTABLE_READY | G_PARAM_READWRITE |
               G_PARAM_STATIC_STRINGS)));
diff --git a/subprojects/gst-plugins-bad/sys/nvcodec/gstnvvp9dec.cpp b/subprojects/gst-plugins-bad/sys/nvcodec/gstnvvp9dec.cpp
index 9280c929bb..04ca0a7045 100644
--- a/subprojects/gst-plugins-bad/sys/nvcodec/gstnvvp9dec.cpp
+++ b/subprojects/gst-plugins-bad/sys/nvcodec/gstnvvp9dec.cpp
@@ -79,7 +79,7 @@ enum
   PROP_MAX_DISPLAY_DELAY,
 };
 
-#define DEFAULT_NUM_OUTPUT_SURFACES 0
+#define DEFAULT_NUM_OUTPUT_SURFACES 1
 #define DEFAULT_MAX_DISPLAY_DELAY -1
 
 static GTypeClass *parent_class = nullptr;
@@ -151,16 +151,17 @@ gst_nv_vp9_dec_class_init (GstNvVp9DecClass * klass,
   /**
    * GstNvVp9Dec:num-output-surfaces:
    *
-   * The number of output surfaces (0 = auto). This property will be used to
-   * calculate the CUVIDDECODECREATEINFO.ulNumOutputSurfaces parameter
-   * in case of CUDA output mode
+   * The number of output surfaces (0 = auto, 1 = always copy).
+   * This property will be used to calculate
+   * the CUVIDDECODECREATEINFO.ulNumOutputSurfaces parameter in case of
+   * CUDA output mode.
    *
    * Since: 1.24
    */
   g_object_class_install_property (object_class, PROP_NUM_OUTPUT_SURFACES,
       g_param_spec_uint ("num-output-surfaces", "Num Output Surfaces",
           "Maximum number of output surfaces simultaneously mapped in CUDA "
-          "output mode (0 = auto)",
+          "output mode (0 = auto, 1 = always copy)",
           0, 64, DEFAULT_NUM_OUTPUT_SURFACES,
           (GParamFlags) (GST_PARAM_MUTABLE_READY | G_PARAM_READWRITE |
               G_PARAM_STATIC_STRINGS)));