diff --git a/dda.c b/dda.c
index be4308708..44b5a6863 100644
--- a/dda.c
+++ b/dda.c
@@ -116,7 +116,10 @@ void dda_new_startpoint(void) {
   startpoint_steps.axis[E] = um_to_steps(startpoint.axis[E], E);
 }
 
-/*! CREATE a dda given current_position and a target, save to passed location so we can write directly into the queue
+/**
+  Create a DDA using startpoint, startpoint_steps and a target, save to passed
+  location so we can write directly into the queue.
+
 	\param *dda pointer to a dda_queue entry to overwrite
 	\param *target the target position of this move
 
@@ -126,8 +129,9 @@ void dda_new_startpoint(void) {
 
 	It also pre-fills any data that the selected accleration algorithm needs, and can be pre-computed for the whole move.
 
-	This algorithm is probably the main limiting factor to print speed in terms of firmware limitations
- *
+  This algorithm is the main limiting factor when queuing movements and can
+  become a limitation to print speed if there are lots of tiny, fast movements.
+
  * Regarding lookahead, we can distinguish everything into these cases:
  *
  * 1. Standard movement. To be joined with the previous move.
@@ -278,6 +282,7 @@ void dda_create(DDA *dda, const TARGET *target) {
 
 	if (dda->total_steps == 0) {
 		dda->nullmove = 1;
+    startpoint.F = dda->endpoint.F;
 	}
 	else {
 		// get steppers ready to go
@@ -467,16 +472,16 @@ void dda_create(DDA *dda, const TARGET *target) {
       if (dda->c < c_limit)
         dda->c = c_limit;
 		#endif
+
+    // next dda starts where we finish
+    memcpy(&startpoint, &dda->endpoint, sizeof(TARGET));
+    #ifdef LOOKAHEAD
+      prev_dda = dda;
+    #endif
 	} /* ! dda->total_steps == 0 */
 
 	if (DEBUG_DDA && (debug_flags & DEBUG_DDA))
 		serial_writestr_P(PSTR("] }\n"));
-
-	// next dda starts where we finish
-	memcpy(&startpoint, &dda->endpoint, sizeof(TARGET));
-  #ifdef LOOKAHEAD
-    prev_dda = dda;
-  #endif
 }
 
 /*! Start a prepared DDA
@@ -498,7 +503,6 @@ void dda_start(DDA *dda) {
                dda->endpoint.axis[X], dda->endpoint.axis[Y],
                dda->endpoint.axis[Z], dda->endpoint.F);
 
-	if ( ! dda->nullmove) {
 		// get ready to go
 		psu_timeout = 0;
     #ifdef Z_AUTODISABLE
@@ -538,10 +542,6 @@ void dda_start(DDA *dda) {
 		// set timeout for first step
     timer_set(dda->c, 0);
 	}
-	// else just a speed change, keep dda->live = 0
-
-	current_position.F = dda->endpoint.F;
-}
 
 /**
   \brief Do per-step movement maintenance.
@@ -971,11 +971,13 @@ void update_current_position() {
       current_position.axis[E] = axis_um;
     }
 
-		// current_position.F is updated in dda_start()
+    current_position.F = dda->endpoint.F;
 	}
   else {
     for (i = X; i < AXIS_COUNT; i++) {
       current_position.axis[i] = startpoint.axis[i];
     }
+    current_position.F = startpoint.F;
 	}
+
 }
diff --git a/dda_lookahead.c b/dda_lookahead.c
index 8b8f7a464..d0ffbd8bf 100644
--- a/dda_lookahead.c
+++ b/dda_lookahead.c
@@ -56,8 +56,8 @@ void dda_find_crossing_speed(DDA *prev, DDA *current) {
   axes_int32_t prevF, currF;
   enum axis_e i;
 
-  // Bail out if there's nothing to join (e.g. G1 F1500).
-  if ( ! prev || prev->nullmove)
+  // Bail out if there's nothing to join (e.g. first movement after a pause).
+  if ( ! prev)
     return;
 
   // We always look at the smaller of both combined speeds,
@@ -180,8 +180,8 @@ void dda_join_moves(DDA *prev, DDA *current) {
   moveno++;
   #endif
 
-  // Bail out if there's nothing to join (e.g. G1 F1500).
-  if ( ! prev || prev->nullmove || current->crossF == 0)
+  // Bail out if there's nothing to join.
+  if ( ! prev || current->crossF == 0)
     return;
 
     // Show the proposed crossing speed - this might get adjusted below.
diff --git a/dda_queue.c b/dda_queue.c
index 54efca9d3..9e619bc34 100644
--- a/dda_queue.c
+++ b/dda_queue.c
@@ -63,7 +63,7 @@ DDA *queue_current_movement() {
   ATOMIC_START
     current = &movebuffer[mb_tail];
 
-    if ( ! current->live || current->waitfor_temp || current->nullmove)
+    if ( ! current->live || current->waitfor_temp)
       current = NULL;
   ATOMIC_END
 
@@ -125,23 +125,32 @@ void enqueue_home(TARGET *t, uint8_t endstop_check, uint8_t endstop_stop_cond) {
 	}
   dda_create(new_movebuffer, t);
 
-	// make certain all writes to global memory
-	// are flushed before modifying mb_head.
-	MEMORY_BARRIER();
+  /**
+    It's pointless to queue up movements which don't actually move the stepper,
+    e.g. pure velocity changes or movements shorter than a single motor step.
 
-	mb_head = h;
+    That said, accept movements which do move the steppers by forwarding
+    mb_head. Also kick off movements if it's the first movement after a pause.
+  */
+  if ( ! new_movebuffer->nullmove) {
+    // make certain all writes to global memory
+    // are flushed before modifying mb_head.
+    MEMORY_BARRIER();
 
-  uint8_t isdead;
+    mb_head = h;
 
-  ATOMIC_START
-    isdead = (movebuffer[mb_tail].live == 0);
-  ATOMIC_END
+    uint8_t isdead;
 
-	if (isdead) {
-    timer_reset();
-		next_move();
-    // Compensate for the cli() in timer_set().
-		sei();
+    ATOMIC_START
+      isdead = (movebuffer[mb_tail].live == 0);
+    ATOMIC_END
+
+    if (isdead) {
+      timer_reset();
+      next_move();
+      // Compensate for the cli() in timer_set().
+      sei();
+    }
 	}
 }
 
@@ -154,7 +163,8 @@ void enqueue_home(TARGET *t, uint8_t endstop_check, uint8_t endstop_stop_cond) {
 /// move buffer was dead in the non-interrupt case (which indicates that the
 /// timer interrupt is disabled).
 void next_move() {
-	while ((queue_empty() == 0) && (movebuffer[mb_tail].live == 0)) {
+
+  if (queue_empty() == 0) {
     // Tail must be set before calling timer_set(), as timer_set() reenables
     // the timer interrupt, potentially exposing mb_tail to the timer
     // interrupt routine.