diff --git a/include/sway/commands.h b/include/sway/commands.h
index 5d468a421..45b5b0f4c 100644
--- a/include/sway/commands.h
+++ b/include/sway/commands.h
@@ -145,6 +145,7 @@ sway_cmd cmd_kill;
 sway_cmd cmd_layout;
 sway_cmd cmd_log_colors;
 sway_cmd cmd_mark;
+sway_cmd cmd_max_render_time;
 sway_cmd cmd_mode;
 sway_cmd cmd_mouse_warping;
 sway_cmd cmd_move;
diff --git a/include/sway/output.h b/include/sway/output.h
index 741f5b5e7..ddc080229 100644
--- a/include/sway/output.h
+++ b/include/sway/output.h
@@ -72,7 +72,7 @@ struct sway_output *output_get_in_direction(struct sway_output *reference,
 void output_add_workspace(struct sway_output *output,
 		struct sway_workspace *workspace);
 
-typedef void (*sway_surface_iterator_func_t)(struct sway_output *output,
+typedef void (*sway_surface_iterator_func_t)(struct sway_output *output, struct sway_view *view,
 	struct wlr_surface *surface, struct wlr_box *box, float rotation,
 	void *user_data);
 
diff --git a/include/sway/surface.h b/include/sway/surface.h
index 06874af2d..4da96c027 100644
--- a/include/sway/surface.h
+++ b/include/sway/surface.h
@@ -6,6 +6,13 @@ struct sway_surface {
 	struct wlr_surface *wlr_surface;
 
 	struct wl_listener destroy;
+
+	/**
+	 * This timer can be used for issuing delayed frame done callbacks (for
+	 * example, to improve presentation latency). Its handler is set to a
+	 * function that issues a frame done callback to this surface.
+	 */
+	struct wl_event_source *frame_done_timer;
 };
 
 #endif
diff --git a/include/sway/tree/view.h b/include/sway/tree/view.h
index 44cd4a7bc..29c87967e 100644
--- a/include/sway/tree/view.h
+++ b/include/sway/tree/view.h
@@ -108,6 +108,8 @@ struct sway_view {
 	} events;
 
 	struct wl_listener surface_new_subsurface;
+
+	int max_render_time; // In milliseconds
 };
 
 struct sway_xdg_shell_view {
diff --git a/sway/commands.c b/sway/commands.c
index e2c43e9fe..e7f1eafe7 100644
--- a/sway/commands.c
+++ b/sway/commands.c
@@ -119,6 +119,7 @@ static struct cmd_handler command_handlers[] = {
 	{ "kill", cmd_kill },
 	{ "layout", cmd_layout },
 	{ "mark", cmd_mark },
+	{ "max_render_time", cmd_max_render_time },
 	{ "move", cmd_move },
 	{ "nop", cmd_nop },
 	{ "opacity", cmd_opacity },
diff --git a/sway/commands/max_render_time.c b/sway/commands/max_render_time.c
new file mode 100644
index 000000000..ec3282f18
--- /dev/null
+++ b/sway/commands/max_render_time.c
@@ -0,0 +1,32 @@
+#include <strings.h>
+#include "sway/commands.h"
+#include "sway/config.h"
+#include "sway/tree/view.h"
+
+struct cmd_results *cmd_max_render_time(int argc, char **argv) {
+	if (!argc) {
+		return cmd_results_new(CMD_INVALID, "Missing max render time argument.");
+	}
+
+	int max_render_time;
+	if (!strcmp(*argv, "off")) {
+		max_render_time = 0;
+	} else {
+		char *end;
+		max_render_time = strtol(*argv, &end, 10);
+		if (*end || max_render_time <= 0) {
+			return cmd_results_new(CMD_INVALID, "Invalid max render time.");
+		}
+	}
+
+	struct sway_container *container = config->handler_context.container;
+	if (!container || !container->view) {
+		return cmd_results_new(CMD_INVALID,
+				"Only views can have a max_render_time");
+	}
+
+	struct sway_view *view = container->view;
+	view->max_render_time = max_render_time;
+
+	return cmd_results_new(CMD_SUCCESS, NULL);
+}
diff --git a/sway/desktop/output.c b/sway/desktop/output.c
index 0f715c19d..6498198b9 100644
--- a/sway/desktop/output.c
+++ b/sway/desktop/output.c
@@ -23,6 +23,7 @@
 #include "sway/layers.h"
 #include "sway/output.h"
 #include "sway/server.h"
+#include "sway/surface.h"
 #include "sway/tree/arrange.h"
 #include "sway/tree/container.h"
 #include "sway/tree/root.h"
@@ -80,6 +81,7 @@ struct surface_iterator_data {
 	void *user_data;
 
 	struct sway_output *output;
+	struct sway_view *view;
 	double ox, oy;
 	int width, height;
 	float rotation;
@@ -134,7 +136,7 @@ static void output_for_each_surface_iterator(struct wlr_surface *surface,
 		return;
 	}
 
-	data->user_iterator(data->output, surface, &box, data->rotation,
+	data->user_iterator(data->output, data->view, surface, &box, data->rotation,
 		data->user_data);
 }
 
@@ -145,6 +147,7 @@ void output_surface_for_each_surface(struct sway_output *output,
 		.user_iterator = iterator,
 		.user_data = user_data,
 		.output = output,
+		.view = NULL,
 		.ox = ox,
 		.oy = oy,
 		.width = surface->current.width,
@@ -163,6 +166,7 @@ void output_view_for_each_surface(struct sway_output *output,
 		.user_iterator = iterator,
 		.user_data = user_data,
 		.output = output,
+		.view = view,
 		.ox = view->container->surface_x - output->lx
 			- view->geometry.x,
 		.oy = view->container->surface_y - output->ly
@@ -182,6 +186,7 @@ void output_view_for_each_popup(struct sway_output *output,
 		.user_iterator = iterator,
 		.user_data = user_data,
 		.output = output,
+		.view = view,
 		.ox = view->container->surface_x - output->lx
 			- view->geometry.x,
 		.oy = view->container->surface_y - output->ly
@@ -224,6 +229,7 @@ void output_layer_for_each_surface(struct sway_output *output,
 				.user_iterator = iterator,
 				.user_data = user_data,
 				.output = output,
+				.view = NULL,
 				.ox = popup_sx,
 				.oy = popup_sy,
 				.width = surface->current.width,
@@ -291,6 +297,7 @@ static void output_for_each_surface(struct sway_output *output,
 		.user_iterator = iterator,
 		.user_data = user_data,
 		.output = output,
+		.view = NULL,
 	};
 
 	struct sway_workspace *workspace = output_get_active_workspace(output);
@@ -401,18 +408,37 @@ bool output_has_opaque_overlay_layer_surface(struct sway_output *output) {
 	return false;
 }
 
-static void send_frame_done_iterator(struct sway_output *output,
+struct send_frame_done_data {
+	struct timespec when;
+	int msec_until_refresh;
+};
+
+static void send_frame_done_iterator(struct sway_output *output, struct sway_view *view,
 		struct wlr_surface *surface, struct wlr_box *box, float rotation,
-		void *data) {
-	struct timespec *when = data;
-	wlr_surface_send_frame_done(surface, when);
+		void *user_data) {
+	int view_max_render_time = 0;
+	if (view != NULL) {
+		view_max_render_time = view->max_render_time;
+	}
+
+	struct send_frame_done_data *data = user_data;
+
+	int delay = data->msec_until_refresh - output->max_render_time
+			- view_max_render_time;
+
+	if (output->max_render_time == 0 || view_max_render_time == 0 || delay < 1) {
+		wlr_surface_send_frame_done(surface, &data->when);
+	} else {
+		struct sway_surface *sway_surface = surface->data;
+		wl_event_source_timer_update(sway_surface->frame_done_timer, delay);
+	}
 }
 
-static void send_frame_done(struct sway_output *output, struct timespec *when) {
-	output_for_each_surface(output, send_frame_done_iterator, when);
+static void send_frame_done(struct sway_output *output, struct send_frame_done_data *data) {
+	output_for_each_surface(output, send_frame_done_iterator, data);
 }
 
-static void count_surface_iterator(struct sway_output *output,
+static void count_surface_iterator(struct sway_output *output, struct sway_view *view,
 		struct wlr_surface *surface, struct wlr_box *_box, float rotation,
 		void *data) {
 	size_t *n = data;
@@ -533,7 +559,7 @@ int output_repaint_timer_handler(void *data) {
 	return 0;
 }
 
-static void damage_handle_frame(struct wl_listener *listener, void *data) {
+static void damage_handle_frame(struct wl_listener *listener, void *user_data) {
 	struct sway_output *output =
 		wl_container_of(listener, output, damage_frame);
 	if (!output->enabled || !output->wlr_output->enabled) {
@@ -592,9 +618,10 @@ static void damage_handle_frame(struct wl_listener *listener, void *data) {
 	}
 
 	// Send frame done to all visible surfaces
-	struct timespec now;
-	clock_gettime(CLOCK_MONOTONIC, &now);
-	send_frame_done(output, &now);
+	struct send_frame_done_data data = {0};
+	clock_gettime(CLOCK_MONOTONIC, &data.when);
+	data.msec_until_refresh = msec_until_refresh;
+	send_frame_done(output, &data);
 }
 
 void output_damage_whole(struct sway_output *output) {
@@ -605,7 +632,7 @@ void output_damage_whole(struct sway_output *output) {
 	}
 }
 
-static void damage_surface_iterator(struct sway_output *output,
+static void damage_surface_iterator(struct sway_output *output, struct sway_view *view,
 		struct wlr_surface *surface, struct wlr_box *_box, float rotation,
 		void *_data) {
 	bool *data = _data;
@@ -811,7 +838,7 @@ static void handle_scale(struct wl_listener *listener, void *data) {
 	update_output_manager_config(output->server);
 }
 
-static void send_presented_iterator(struct sway_output *output,
+static void send_presented_iterator(struct sway_output *output, struct sway_view *view,
 		struct wlr_surface *surface, struct wlr_box *box, float rotation,
 		void *data) {
 	struct wlr_presentation_event *event = data;
diff --git a/sway/desktop/render.c b/sway/desktop/render.c
index 916d4eba1..c432b4769 100644
--- a/sway/desktop/render.c
+++ b/sway/desktop/render.c
@@ -97,7 +97,7 @@ damage_finish:
 	pixman_region32_fini(&damage);
 }
 
-static void render_surface_iterator(struct sway_output *output,
+static void render_surface_iterator(struct sway_output *output, struct sway_view *view,
 		struct wlr_surface *surface, struct wlr_box *_box, float rotation,
 		void *_data) {
 	struct render_data *data = _data;
@@ -214,11 +214,11 @@ static void render_view_toplevels(struct sway_view *view,
 			render_surface_iterator, &data);
 }
 
-static void render_popup_iterator(struct sway_output *output,
+static void render_popup_iterator(struct sway_output *output, struct sway_view *view,
 		struct wlr_surface *surface, struct wlr_box *box, float rotation,
 		void *data) {
 	// Render this popup's surface
-	render_surface_iterator(output, surface, box, rotation, data);
+	render_surface_iterator(output, view, surface, box, rotation, data);
 
 	// Render this popup's child toplevels
 	output_surface_for_each_surface(output, surface, box->x, box->y,
diff --git a/sway/desktop/surface.c b/sway/desktop/surface.c
index 41d4ce3fc..853c403d5 100644
--- a/sway/desktop/surface.c
+++ b/sway/desktop/surface.c
@@ -1,4 +1,6 @@
+#define _POSIX_C_SOURCE 200112L
 #include <stdlib.h>
+#include <time.h>
 #include <wlr/types/wlr_surface.h>
 #include "sway/server.h"
 #include "sway/surface.h"
@@ -9,9 +11,21 @@ void handle_destroy(struct wl_listener *listener, void *data) {
 	surface->wlr_surface->data = NULL;
 	wl_list_remove(&surface->destroy.link);
 
+	wl_event_source_remove(surface->frame_done_timer);
+
 	free(surface);
 }
 
+static int surface_frame_done_timer_handler(void *data) {
+	struct sway_surface *surface = data;
+
+	struct timespec now;
+	clock_gettime(CLOCK_MONOTONIC, &now);
+	wlr_surface_send_frame_done(surface->wlr_surface, &now);
+
+	return 0;
+}
+
 void handle_compositor_new_surface(struct wl_listener *listener, void *data) {
 	struct wlr_surface *wlr_surface = data;
 
@@ -21,4 +35,7 @@ void handle_compositor_new_surface(struct wl_listener *listener, void *data) {
 
 	surface->destroy.notify = handle_destroy;
 	wl_signal_add(&wlr_surface->events.destroy, &surface->destroy);
+
+	surface->frame_done_timer = wl_event_loop_add_timer(server.wl_event_loop,
+		surface_frame_done_timer_handler, surface);
 }
diff --git a/sway/meson.build b/sway/meson.build
index 125401545..e285c09ed 100644
--- a/sway/meson.build
+++ b/sway/meson.build
@@ -68,6 +68,7 @@ sway_sources = files(
 	'commands/inhibit_idle.c',
 	'commands/kill.c',
 	'commands/mark.c',
+	'commands/max_render_time.c',
 	'commands/opacity.c',
 	'commands/include.c',
 	'commands/input.c',
diff --git a/sway/sway-output.5.scd b/sway/sway-output.5.scd
index 01496f194..858e73c10 100644
--- a/sway/sway-output.5.scd
+++ b/sway/sway-output.5.scd
@@ -131,6 +131,9 @@ must be separated by one space. For example:
 	. Start with *max_render_time 1*. Increment by *1* if you see frame
 	  drops.
 
+	To achieve even lower latency, see the *max_render_time* surface
+	property in *sway*(5).
+
 # SEE ALSO
 
 *sway*(5) *sway-input*(5)
diff --git a/sway/sway.5.scd b/sway/sway.5.scd
index 9119b379a..52ee9d28e 100644
--- a/sway/sway.5.scd
+++ b/sway/sway.5.scd
@@ -182,6 +182,22 @@ set|plus|minus <amount>
 *layout* toggle [split|tabbed|stacking|splitv|splith] [split|tabbed|stacking|splitv|splith]...
 	Cycles the layout mode of the focused container through a list of layouts.
 
+*max_render_time* off|<msec>
+	Works together with *output max_render_time* to reduce the latency even
+	further by delaying the frame callbacks sent to a surface. When set to
+	a positive number of milliseconds, delays the frame callback in such a
+	way that the surface has the specified number of milliseconds to render
+	and commit new contents before being sampled by the compositor for the
+	next presentation. See *max_render_time* in *sway-output*(5) for
+	further details.
+
+	To set this up for optimal latency:
+	. Set up *output max_render_time*.
+	. Put the target application in _full-screen_ and have it continuously
+	  render something.
+	. Start by setting *max_render_time 1*. If the application drops
+	  frames, increment by *1*.
+
 *move* left|right|up|down [<px> px]
 	Moves the focused container in the direction specified. If the container,
 	the optional _px_ argument specifies how many pixels to move the container.