diff --git a/CMakeLists.txt b/CMakeLists.txt
index 951160a3b..e0110c69f 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -612,6 +612,7 @@ foreach(
   SDL1_VIDEO_MODE_BPP
   SDL1_VIDEO_MODE_FLAGS
   SDL1_FORCE_SVID_VIDEO_MODE
+  SDL1_FORCE_DIRECT_RENDER
   HAS_KBCTRL
   KBCTRL_BUTTON_DPAD_LEFT
   KBCTRL_BUTTON_DPAD_RIGHT
diff --git a/Source/dx.cpp b/Source/dx.cpp
index 255cc6911..a49443e21 100644
--- a/Source/dx.cpp
+++ b/Source/dx.cpp
@@ -39,16 +39,45 @@ SDL_Surface *renderer_texture_surface = nullptr;
 /** 8-bit surface that we render to */
 SDL_Surface *pal_surface;
 
+/** Whether we render directly to the screen surface, i.e. `pal_surface == GetOutputSurface()` */
+bool RenderDirectlyToOutputSurface;
+
+namespace {
+
+bool CanRenderDirectlyToOutputSurface()
+{
+#ifdef USE_SDL1
+#ifdef SDL1_FORCE_DIRECT_RENDER
+	return true;
+#else
+	auto *outputSurface = GetOutputSurface();
+	return ((outputSurface->flags & SDL_DOUBLEBUF) == SDL_DOUBLEBUF
+	    && outputSurface->w == gnScreenWidth && outputSurface->h == gnScreenHeight
+	    && outputSurface->format->BitsPerPixel == 8);
+#endif
+#else // !USE_SDL1
+	return false;
+#endif
+}
+
+} // namespace
+
 static void dx_create_back_buffer()
 {
-	pal_surface = SDL_CreateRGBSurfaceWithFormat(
-	    /*flags=*/0,
-	    /*width=*/gnScreenWidth,
-	    /*height=*/gnScreenHeight,
-	    /*depth=*/8,
-	    SDL_PIXELFORMAT_INDEX8);
-	if (pal_surface == nullptr) {
-		ErrSdl();
+	if (CanRenderDirectlyToOutputSurface()) {
+		Log("{}", "Will render directly to the SDL output surface");
+		pal_surface = GetOutputSurface();
+		RenderDirectlyToOutputSurface = true;
+	} else {
+		pal_surface = SDL_CreateRGBSurfaceWithFormat(
+		    /*flags=*/0,
+		    /*width=*/gnScreenWidth,
+		    /*height=*/gnScreenHeight,
+		    /*depth=*/8,
+		    SDL_PIXELFORMAT_INDEX8);
+		if (pal_surface == nullptr) {
+			ErrSdl();
+		}
 	}
 
 #ifndef USE_SDL1
@@ -194,6 +223,8 @@ void InitPalette()
 
 void BltFast(SDL_Rect *src_rect, SDL_Rect *dst_rect)
 {
+	if (RenderDirectlyToOutputSurface)
+		return;
 	Blit(pal_surface, src_rect, dst_rect);
 }
 
@@ -315,6 +346,8 @@ void RenderPresent()
 	if (SDL_Flip(surface) <= -1) {
 		ErrSdl();
 	}
+	if (RenderDirectlyToOutputSurface)
+		pal_surface = GetOutputSurface();
 	LimitFrameRate();
 #endif
 }
diff --git a/Source/dx.h b/Source/dx.h
index a52da3db9..1c2dbbad5 100644
--- a/Source/dx.h
+++ b/Source/dx.h
@@ -9,6 +9,9 @@
 
 namespace devilution {
 
+/** Whether we render directly to the screen surface, i.e. `pal_surface == GetOutputSurface()` */
+extern bool RenderDirectlyToOutputSurface;
+
 CelOutputBuffer GlobalBackBuffer();
 
 void dx_init();
diff --git a/Source/scrollrt.cpp b/Source/scrollrt.cpp
index 96177336a..a9dea66fe 100644
--- a/Source/scrollrt.cpp
+++ b/Source/scrollrt.cpp
@@ -143,7 +143,7 @@ static void BlitCursor(BYTE *dst, int dst_pitch, BYTE *src, int src_pitch)
 /**
  * @brief Remove the cursor from the buffer
  */
-static void scrollrt_draw_cursor_back_buffer(const CelOutputBuffer &out)
+static void UndrawCursor(const CelOutputBuffer &out)
 {
 	if (sgdwCursWdt == 0) {
 		return;
@@ -159,12 +159,10 @@ static void scrollrt_draw_cursor_back_buffer(const CelOutputBuffer &out)
 }
 
 /**
- * @brief Draw the cursor on the given buffer
+ * @brief Save the content behind the cursor to a temporary buffer, then draw the cursor.
  */
-static void scrollrt_draw_cursor_item(const CelOutputBuffer &out)
+static void DrawCursor(const CelOutputBuffer &out)
 {
-	assert(!sgdwCursWdt);
-
 	if (pcurs <= CURSOR_NONE || cursW == 0 || cursH == 0) {
 		return;
 	}
@@ -1426,7 +1424,7 @@ static void DoBlitScreen(Sint16 dwX, Sint16 dwY, Uint16 dwWdt, Uint16 dwHgt)
  */
 static void DrawMain(int dwHgt, bool draw_desc, bool draw_hp, bool draw_mana, bool draw_sbar, bool draw_btn)
 {
-	if (!gbActive) {
+	if (!gbActive || RenderDirectlyToOutputSurface) {
 		return;
 	}
 
@@ -1481,18 +1479,19 @@ void scrollrt_draw_game_screen(bool draw_cursor)
 
 	if (draw_cursor) {
 		lock_buf(0);
-		scrollrt_draw_cursor_item(GlobalBackBuffer());
+		DrawCursor(GlobalBackBuffer());
 		unlock_buf(0);
 	}
 
 	DrawMain(hgt, false, false, false, false, false);
 
+	RenderPresent();
+
 	if (draw_cursor) {
 		lock_buf(0);
-		scrollrt_draw_cursor_back_buffer(GlobalBackBuffer());
+		UndrawCursor(GlobalBackBuffer());
 		unlock_buf(0);
 	}
-	RenderPresent();
 }
 
 /**
@@ -1526,6 +1525,7 @@ void DrawAndBlit()
 
 	lock_buf(0);
 	const CelOutputBuffer &out = GlobalBackBuffer();
+	UndrawCursor(out);
 
 	nthread_UpdateProgressToNextGameTick();
 
@@ -1550,7 +1550,7 @@ void DrawAndBlit()
 		hgt = gnScreenHeight;
 	}
 	DrawXPBar(out);
-	scrollrt_draw_cursor_item(out);
+	DrawCursor(out);
 
 	DrawFPS(out);
 
@@ -1558,9 +1558,6 @@ void DrawAndBlit()
 
 	DrawMain(hgt, ddsdesc, drawhpflag, drawmanaflag, drawsbarflag, drawbtnflag);
 
-	lock_buf(0);
-	scrollrt_draw_cursor_back_buffer(GlobalBackBuffer());
-	unlock_buf(0);
 	RenderPresent();
 
 	drawhpflag = false;
diff --git a/Source/utils/display.cpp b/Source/utils/display.cpp
index fe718e01e..023ebd902 100644
--- a/Source/utils/display.cpp
+++ b/Source/utils/display.cpp
@@ -39,11 +39,13 @@ Uint16 gnViewportHeight;
 void SetVideoMode(int width, int height, int bpp, uint32_t flags)
 {
 	Log("Setting video mode {}x{} bpp={} flags=0x{:08X}", width, height, bpp, flags);
-	SDL_SetVideoMode(width, height, bpp, flags);
+	ghMainWnd = SDL_SetVideoMode(width, height, bpp, flags);
+	if (ghMainWnd == nullptr) {
+		ErrSdl();
+	}
 	const SDL_VideoInfo &current = *SDL_GetVideoInfo();
 	Log("Video mode is now {}x{} bpp={} flags=0x{:08X}",
 	    current.current_w, current.current_h, current.vfmt->BitsPerPixel, SDL_GetVideoSurface()->flags);
-	ghMainWnd = SDL_GetVideoSurface();
 }
 
 void SetVideoModeToPrimary(bool fullscreen, int width, int height)