diff --git a/Makefile b/Makefile
index 7baff23a57..58212f2bec 100644
--- a/Makefile
+++ b/Makefile
@@ -77,9 +77,8 @@ ifneq ($(findstring Linux,$(OS)),)
    JOYCONFIG_OBJ += input/linuxraw_joypad.o
 endif
 
-OBJ += autosave.o thread.o
-
 ifeq ($(HAVE_THREADS), 1)
+   OBJ += autosave.o thread.o gfx/thread_wrapper.o
    ifeq ($(findstring Haiku,$(OS)),)
       LIBS += -lpthread
    endif
diff --git a/Makefile.win b/Makefile.win
index 79396f99bc..a038dece6a 100644
--- a/Makefile.win
+++ b/Makefile.win
@@ -99,7 +99,7 @@ ifeq ($(HAVE_SDL), 1)
 endif
 
 ifeq ($(HAVE_THREADS), 1)
-   OBJ += autosave.o thread.o
+   OBJ += autosave.o thread.o gfx/thread_wrapper.o
    DEFINES += -DHAVE_THREADS
 endif
 
diff --git a/config.def.h b/config.def.h
index 7a0cf21cac..779117199b 100644
--- a/config.def.h
+++ b/config.def.h
@@ -218,6 +218,9 @@ static const bool disable_composition = false;
 // Video VSYNC (recommended)
 static const bool vsync = true;
 
+// Threaded video. Will possibly increase performance significantly at cost of worse synchronization and latency.
+static const bool video_threaded = false;
+
 // Smooths picture
 static const bool video_smooth = true;
 
diff --git a/driver.c b/driver.c
index 364ecba733..5db2e09af8 100644
--- a/driver.c
+++ b/driver.c
@@ -23,6 +23,7 @@
 #include "compat/posix_string.h"
 #include "audio/utils.h"
 #include "audio/resampler.h"
+#include "gfx/thread_wrapper.h"
 
 #ifdef HAVE_X11
 #include "gfx/context/x11_common.h"
@@ -492,6 +493,12 @@ static void compute_audio_buffer_statistics(void)
 
 static void compute_monitor_fps_statistics(void)
 {
+   if (g_settings.video.threaded)
+   {
+      RARCH_LOG("Monitor FPS estimation is disabled for threaded video.\n");
+      return;
+   }
+
    if (g_extern.measure_data.frame_time_samples_count < 2 * MEASURE_FRAME_TIME_SAMPLES_COUNT)
    {
       RARCH_LOG("Does not have enough samples for monitor refresh rate estimation. Requires to run for at least %u frames.\n",
@@ -800,7 +807,22 @@ void init_video_input(void)
    video.rgb32 = g_extern.filter.active || (g_extern.system.pix_fmt == RETRO_PIXEL_FORMAT_XRGB8888);
 
    const input_driver_t *tmp = driver.input;
-   driver.video_data = video_init_func(&video, &driver.input, &driver.input_data);
+#ifdef HAVE_THREADS
+   if (g_settings.video.threaded)
+   {
+      find_video_driver(); // Need to grab the "real" video driver interface on a reinit.
+      RARCH_LOG("Starting threaded video driver ...\n");
+      if (!rarch_threaded_video_init(&driver.video, &driver.video_data,
+               &driver.input, &driver.input_data,
+               driver.video, &video))
+      {
+         RARCH_ERR("Cannot open threaded video driver ... Exiting ...\n");
+         rarch_fail(1, "init_video_input()");
+      }
+   }
+   else
+#endif
+      driver.video_data = video_init_func(&video, &driver.input, &driver.input_data);
 
    if (driver.video_data == NULL)
    {
diff --git a/driver.h b/driver.h
index 580999e379..59c06b63ee 100644
--- a/driver.h
+++ b/driver.h
@@ -280,6 +280,8 @@ typedef struct driver
    void *video_data;
    void *input_data;
 
+   bool threaded_video;
+
    // Set if the respective handles are owned by RetroArch driver core.
    // Consoles upper logic will generally intialize the drivers before
    // the driver core initializes. It will then be up to upper logic
diff --git a/general.h b/general.h
index e6bfa575fe..4398447637 100644
--- a/general.h
+++ b/general.h
@@ -173,6 +173,7 @@ struct settings
       char filter_path[PATH_MAX];
       enum rarch_shader_type shader_type;
       float refresh_rate;
+      bool threaded;
 
       bool render_to_texture;
 
diff --git a/gfx/d3d9/d3d9.cpp b/gfx/d3d9/d3d9.cpp
index 70efc7a69e..856838f6d3 100644
--- a/gfx/d3d9/d3d9.cpp
+++ b/gfx/d3d9/d3d9.cpp
@@ -1160,9 +1160,12 @@ static void *d3d9_init(const video_info_t *info, const input_driver_t **input,
      if (!vid)
         return nullptr;
 
-     void *dinput = input_dinput.init();
-     *input       = dinput ? &input_dinput : nullptr;
-     *input_data  = dinput;
+     if (input && input_data)
+     {
+        void *dinput = input_dinput.init();
+        *input       = dinput ? &input_dinput : nullptr;
+        *input_data  = dinput;
+     }
 
      return vid;
    }
diff --git a/gfx/gl.c b/gfx/gl.c
index 0ebd219329..ac3a6739b8 100644
--- a/gfx/gl.c
+++ b/gfx/gl.c
@@ -1691,7 +1691,8 @@ static void *gl_init(const video_info_t *video, const input_driver_t **input, vo
    gl_init_textures(gl, video);
    gl_init_textures_data(gl);
 
-   context_input_driver_func(input, input_data);
+   if (input && input_data)
+      context_input_driver_func(input, input_data);
    
 #ifndef HAVE_RMENU
    // Comes too early for console - moved to gl_start
diff --git a/gfx/sdl_gfx.c b/gfx/sdl_gfx.c
index 0f30ad2778..2b2319af86 100644
--- a/gfx/sdl_gfx.c
+++ b/gfx/sdl_gfx.c
@@ -231,16 +231,19 @@ static void *sdl_gfx_init(const video_info_t *video, const input_driver_t **inpu
 
    sdl_gfx_set_handles();
 
-   sdl_input = input_sdl.init();
-   if (sdl_input)
+   if (input && input_data)
    {
-      *input = &input_sdl;
-      *input_data = sdl_input;
-   }
-   else
-   {
-      *input = NULL;
-      *input_data = NULL;
+      sdl_input = input_sdl.init();
+      if (sdl_input)
+      {
+         *input = &input_sdl;
+         *input_data = sdl_input;
+      }
+      else
+      {
+         *input = NULL;
+         *input_data = NULL;
+      }
    }
 
    sdl_init_font(vid, g_settings.video.font_path, g_settings.video.font_size);
diff --git a/gfx/thread_wrapper.c b/gfx/thread_wrapper.c
new file mode 100644
index 0000000000..6f5337b478
--- /dev/null
+++ b/gfx/thread_wrapper.c
@@ -0,0 +1,316 @@
+/*  RetroArch - A frontend for libretro.
+ *  Copyright (C) 2010-2013 - Hans-Kristian Arntzen
+ * 
+ *  RetroArch is free software: you can redistribute it and/or modify it under the terms
+ *  of the GNU General Public License as published by the Free Software Found-
+ *  ation, either version 3 of the License, or (at your option) any later version.
+ *
+ *  RetroArch is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY;
+ *  without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR
+ *  PURPOSE.  See the GNU General Public License for more details.
+ *
+ *  You should have received a copy of the GNU General Public License along with RetroArch.
+ *  If not, see <http://www.gnu.org/licenses/>.
+ */
+
+#include "thread_wrapper.h"
+#include "../thread.h"
+#include "../general.h"
+#include <stdlib.h>
+#include <string.h>
+#include <limits.h>
+
+enum thread_cmd
+{
+   CMD_NONE = 0,
+   CMD_INIT,
+   CMD_SET_SHADER,
+   CMD_FREE,
+   CMD_SET_ROTATION,
+   CMD_VIEWPORT_INFO,
+   CMD_READ_VIEWPORT,
+   CMD_SET_NONBLOCK,
+
+   CMD_DUMMY = INT_MAX
+};
+
+typedef struct thread_video
+{
+   slock_t *lock;
+   scond_t *cond_cmd;
+   scond_t *cond_thread;
+   sthread_t *thread;
+
+   video_info_t info;
+   const video_driver_t *driver;
+   void *driver_data;
+   const input_driver_t **input;
+   void **input_data;
+
+   bool alive;
+   bool focus;
+
+   enum thread_cmd send_cmd;
+   enum thread_cmd reply_cmd;
+   union
+   {
+      bool b;
+      int i;
+      const char *str;
+      void *v;
+   } cmd_data;
+
+   struct
+   {
+      slock_t *lock;
+      uint8_t *buffer;
+      unsigned width;
+      unsigned height;
+      unsigned pitch;
+      bool updated;
+      char msg[1024];
+   } frame;
+
+} thread_video_t;
+
+static void *thread_init_never_call(const video_info_t *video, const input_driver_t **input, void **input_data)
+{
+   (void)video;
+   (void)input;
+   (void)input_data;
+   RARCH_ERR("Sanity check fail! Threaded mustn't be reinit.\n");
+   abort();
+   return NULL;
+}
+
+static void thread_reply(thread_video_t *thr, enum thread_cmd cmd)
+{
+   slock_lock(thr->lock);
+   thr->reply_cmd = cmd;
+   thr->send_cmd = CMD_NONE;
+   scond_signal(thr->cond_cmd);
+   slock_unlock(thr->lock);
+}
+
+static void thread_loop(void *data)
+{
+   thread_video_t *thr = (thread_video_t*)data;
+
+   for (;;)
+   {
+      bool updated = false;
+      slock_lock(thr->lock);
+      while (thr->send_cmd == CMD_NONE && !thr->frame.updated)
+         scond_wait(thr->cond_thread, thr->lock);
+      if (thr->frame.updated)
+         updated = true;
+      slock_unlock(thr->lock);
+
+      switch (thr->send_cmd)
+      {
+         case CMD_INIT:
+            //fprintf(stderr, "CMD_INIT\n");
+            thr->driver_data = thr->driver->init(&thr->info, thr->input, thr->input_data);
+            thr->cmd_data.b = thr->driver_data;
+            thread_reply(thr, CMD_INIT);
+            break;
+
+         case CMD_FREE:
+            //fprintf(stderr, "CMD_FREE\n");
+            if (thr->driver_data)
+               thr->driver->free(thr->driver_data);
+            thr->driver_data = NULL;
+            thread_reply(thr, CMD_FREE);
+            return;
+
+         case CMD_SET_NONBLOCK:
+            //fprintf(stderr, "CMD_SET_NONBLOCK\n");
+            thr->driver->set_nonblock_state(thr->driver_data, thr->cmd_data.b);
+            thread_reply(thr, CMD_SET_NONBLOCK);
+            break;
+
+         default:
+            //fprintf(stderr, "CMD unknown ...\n");
+            thread_reply(thr, thr->send_cmd);
+            break;
+      }
+
+      if (updated)
+      {
+         //fprintf(stderr, "RUN FRAME\n");
+         slock_lock(thr->frame.lock);
+         bool ret = thr->driver->frame(thr->driver_data,
+               thr->frame.buffer, thr->frame.width, thr->frame.height,
+               thr->frame.pitch, *thr->frame.msg ? thr->frame.msg : NULL);
+         slock_unlock(thr->frame.lock);
+
+         bool alive = ret && thr->driver->alive(thr->driver_data);
+         bool focus = ret && thr->driver->focus(thr->driver_data);
+         //fprintf(stderr, "Alive: %d, Focus: %d.\n", alive, focus);
+
+         slock_lock(thr->lock);
+         thr->alive = alive;
+         thr->focus = focus;
+         thr->frame.updated = false;
+         slock_unlock(thr->lock);
+      }
+   }
+}
+
+static void thread_send_cmd(thread_video_t *thr, enum thread_cmd cmd)
+{
+   slock_lock(thr->lock);
+   thr->send_cmd = cmd;
+   thr->reply_cmd = CMD_NONE;
+   scond_signal(thr->cond_thread);
+   slock_unlock(thr->lock);
+}
+
+static void thread_wait_reply(thread_video_t *thr, enum thread_cmd cmd)
+{
+   slock_lock(thr->lock);
+   while (cmd != thr->reply_cmd)
+      scond_wait(thr->cond_cmd, thr->lock);
+   slock_unlock(thr->lock);
+}
+
+static bool thread_alive(void *data)
+{
+   thread_video_t *thr = (thread_video_t*)data;
+   slock_lock(thr->lock);
+   bool ret = thr->alive;
+   slock_unlock(thr->lock);
+   return ret;
+}
+
+static bool thread_focus(void *data)
+{
+   thread_video_t *thr = (thread_video_t*)data;
+   slock_lock(thr->lock);
+   bool ret = thr->focus;
+   slock_unlock(thr->lock);
+   return ret;
+}
+
+static bool thread_frame(void *data, const void *frame_,
+      unsigned width, unsigned height, unsigned pitch, const char *msg)
+{
+   if (!frame_)
+      return true;
+
+   thread_video_t *thr = (thread_video_t*)data;
+   unsigned copy_stride = width * (thr->info.rgb32 ? sizeof(uint32_t) : sizeof(uint16_t));
+
+   const uint8_t *src = (const uint8_t*)frame_;
+   uint8_t *dst = thr->frame.buffer;
+
+   slock_lock(thr->lock);
+   // Drop frame if updated flag is still set, as thread is still working on last frame.
+   if (!thr->frame.updated)
+   {
+      slock_lock(thr->frame.lock);
+      for (unsigned h = 0; h < height; h++, src += pitch, dst += copy_stride)
+         memcpy(dst, src, copy_stride);
+      thr->frame.updated = true;
+      scond_signal(thr->cond_thread);
+      thr->frame.width  = width;
+      thr->frame.height = height;
+      thr->frame.pitch  = copy_stride;
+
+      if (msg)
+         strlcpy(thr->frame.msg, msg, sizeof(thr->frame.msg));
+      else
+         *thr->frame.msg = '\0';
+
+      slock_unlock(thr->frame.lock);
+   }
+   slock_unlock(thr->lock);
+
+   return true;
+}
+
+static void thread_set_nonblock_state(void *data, bool state)
+{
+   thread_video_t *thr = (thread_video_t*)data;
+   thr->cmd_data.b = state;
+   thread_send_cmd(thr, CMD_SET_NONBLOCK);
+   thread_wait_reply(thr, CMD_SET_NONBLOCK);
+}
+
+static bool thread_init(thread_video_t *thr, const video_info_t *info, const input_driver_t **input,
+      void **input_data)
+{
+   thr->lock = slock_new();
+   thr->frame.lock = slock_new();
+   thr->cond_cmd = scond_new();
+   thr->cond_thread = scond_new();
+   thr->input = input;
+   thr->input_data = input_data;
+   thr->info = *info;
+   thr->alive = true;
+   thr->focus = true;
+
+   thr->frame.buffer = (uint8_t*)malloc((info->rgb32 ? sizeof(uint32_t) : sizeof(uint16_t)) * 
+         info->input_scale * info->input_scale * RARCH_SCALE_BASE * RARCH_SCALE_BASE);
+   if (!thr->frame.buffer)
+      return false;
+
+   thr->thread = sthread_create(thread_loop, thr);
+   if (!thr->thread)
+      return false;
+   thread_send_cmd(thr, CMD_INIT);
+   thread_wait_reply(thr, CMD_INIT);
+   return thr->cmd_data.b;
+}
+
+static void thread_free(void *data)
+{
+   thread_video_t *thr = (thread_video_t*)data;
+   if (!thr)
+      return;
+
+   thread_send_cmd(thr, CMD_FREE);
+   thread_wait_reply(thr, CMD_FREE);
+   sthread_join(thr->thread);
+
+   free(thr->frame.buffer);
+   slock_free(thr->frame.lock);
+   slock_free(thr->lock);
+   scond_free(thr->cond_cmd);
+   scond_free(thr->cond_thread);
+
+   free(thr);
+}
+
+static const video_driver_t video_thread = {
+   thread_init_never_call, // Should never be called directly.
+   thread_frame,
+   thread_set_nonblock_state,
+   thread_alive,
+   thread_focus,
+   NULL, // set_shader
+   thread_free,
+   "Thread wrapper",
+   NULL, // set_rotation
+   NULL, // viewport_info
+   NULL, // read_viewport
+#ifdef HAVE_OVERLAY
+   NULL, // get_overlay_interface
+#endif
+};
+
+bool rarch_threaded_video_init(const video_driver_t **out_driver, void **out_data,
+      const input_driver_t **input, void **input_data,
+      const video_driver_t *driver, const video_info_t *info)
+{
+   thread_video_t *thr = (thread_video_t*)calloc(1, sizeof(*thr));
+   if (!thr)
+      return false;
+
+   thr->driver = driver;
+   *out_driver = &video_thread;
+   *out_data   = thr;
+   return thread_init(thr, info, input, input_data);
+}
+
+
diff --git a/gfx/thread_wrapper.h b/gfx/thread_wrapper.h
new file mode 100644
index 0000000000..6019803525
--- /dev/null
+++ b/gfx/thread_wrapper.h
@@ -0,0 +1,24 @@
+/*  RetroArch - A frontend for libretro.
+ *  Copyright (C) 2010-2013 - Hans-Kristian Arntzen
+ * 
+ *  RetroArch is free software: you can redistribute it and/or modify it under the terms
+ *  of the GNU General Public License as published by the Free Software Found-
+ *  ation, either version 3 of the License, or (at your option) any later version.
+ *
+ *  RetroArch is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY;
+ *  without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR
+ *  PURPOSE.  See the GNU General Public License for more details.
+ *
+ *  You should have received a copy of the GNU General Public License along with RetroArch.
+ *  If not, see <http://www.gnu.org/licenses/>.
+ */
+
+#include "../driver.h"
+#include "../boolean.h"
+
+// Starts a video driver in a new thread.
+// Access to video driver will be mediated through this driver.
+bool rarch_threaded_video_init(const video_driver_t **out_driver, void **out_data,
+      const input_driver_t **input, void **input_data,
+      const video_driver_t *driver, const video_info_t *info);
+
diff --git a/gfx/xvideo.c b/gfx/xvideo.c
index a4db814332..ded1a37cd7 100644
--- a/gfx/xvideo.c
+++ b/gfx/xvideo.c
@@ -464,14 +464,17 @@ static void *xv_init(const video_info_t *video, const input_driver_t **input, vo
    driver.video_display = (uintptr_t)xv->display;
    driver.video_window  = (Window)xv->window;
 
-   xinput = input_x.init();
-   if (xinput)
+   if (input && input_data)
    {
-      *input = &input_x;
-      *input_data = xinput;
+      xinput = input_x.init();
+      if (xinput)
+      {
+         *input = &input_x;
+         *input_data = xinput;
+      }
+      else
+         *input = NULL;
    }
-   else
-      *input = NULL;
 
    init_yuv_tables(xv);
    xv_init_font(xv, g_settings.video.font_path, g_settings.video.font_size);
diff --git a/retroarch.cfg b/retroarch.cfg
index 2237d20e35..22248ce2cf 100644
--- a/retroarch.cfg
+++ b/retroarch.cfg
@@ -65,6 +65,9 @@
 # Video vsync.
 # video_vsync = true
 
+# Use threaded video driver. Using this might improve performance at possible cost of latency and more video stuttering.
+# video_threaded = false
+
 # Smoothens picture with bilinear filtering. Should be disabled if using pixel shaders.
 # video_smooth = true
 
diff --git a/settings.c b/settings.c
index d6a798f20f..217924c87d 100644
--- a/settings.c
+++ b/settings.c
@@ -159,6 +159,7 @@ void config_set_defaults(void)
    g_settings.video.fullscreen_y = fullscreen_y;
    g_settings.video.disable_composition = disable_composition;
    g_settings.video.vsync = vsync;
+   g_settings.video.threaded = video_threaded;
    g_settings.video.smooth = video_smooth;
    g_settings.video.force_aspect = force_aspect;
    g_settings.video.scale_integer = scale_integer;
@@ -438,6 +439,7 @@ bool config_load_file(const char *path)
    CONFIG_GET_INT(video.monitor_index, "video_monitor_index");
    CONFIG_GET_BOOL(video.disable_composition, "video_disable_composition");
    CONFIG_GET_BOOL(video.vsync, "video_vsync");
+   CONFIG_GET_BOOL(video.threaded, "video_threaded");
    CONFIG_GET_BOOL(video.smooth, "video_smooth");
    CONFIG_GET_BOOL(video.force_aspect, "video_force_aspect");
    CONFIG_GET_BOOL(video.scale_integer, "video_scale_integer");