diff --git a/src/backend/executor/nodeIncrementalSort.c b/src/backend/executor/nodeIncrementalSort.c
index 4f50bc845da..d1b97d46bc0 100644
--- a/src/backend/executor/nodeIncrementalSort.c
+++ b/src/backend/executor/nodeIncrementalSort.c
@@ -315,7 +315,7 @@ switchToPresortedPrefixMode(PlanState *pstate)
 												&(plannode->sort.nullsFirst[nPresortedCols]),
 												work_mem,
 												NULL,
-												TUPLESORT_NONE);
+												node->bounded ? TUPLESORT_ALLOWBOUNDED : TUPLESORT_NONE);
 		node->prefixsort_state = prefixsort_state;
 	}
 	else
@@ -616,6 +616,8 @@ ExecIncrementalSort(PlanState *pstate)
 												  plannode->sort.nullsFirst,
 												  work_mem,
 												  NULL,
+												  node->bounded ?
+												  TUPLESORT_ALLOWBOUNDED :
 												  TUPLESORT_NONE);
 			node->fullsort_state = fullsort_state;
 		}
diff --git a/src/backend/executor/nodeSort.c b/src/backend/executor/nodeSort.c
index a113d737955..3c28d60c3ef 100644
--- a/src/backend/executor/nodeSort.c
+++ b/src/backend/executor/nodeSort.c
@@ -99,6 +99,8 @@ ExecSort(PlanState *pstate)
 
 		if (node->randomAccess)
 			tuplesortopts |= TUPLESORT_RANDOMACCESS;
+		if (node->bounded)
+			tuplesortopts |= TUPLESORT_ALLOWBOUNDED;
 
 		if (node->datumSort)
 			tuplesortstate = tuplesort_begin_datum(TupleDescAttr(tupDesc, 0)->atttypid,
diff --git a/src/backend/utils/sort/tuplesort.c b/src/backend/utils/sort/tuplesort.c
index a8a5cc52047..571fb955327 100644
--- a/src/backend/utils/sort/tuplesort.c
+++ b/src/backend/utils/sort/tuplesort.c
@@ -935,11 +935,21 @@ tuplesort_begin_batch(Tuplesortstate *state)
 	 * eases memory management.  Resetting at key points reduces
 	 * fragmentation. Note that the memtuples array of SortTuples is allocated
 	 * in the parent context, not this context, because there is no need to
-	 * free memtuples early.
+	 * free memtuples early.  For bounded sorts, tuples may be pfreed in any
+	 * order, so we use a regular aset.c context so that it can make use of
+	 * free'd memory.  When the sort is not bounded, we make use of a
+	 * generation.c context as this keeps allocations more compact with less
+	 * wastage.  Allocations are also slightly more CPU efficient.
 	 */
-	state->tuplecontext = AllocSetContextCreate(state->sortcontext,
-												"Caller tuples",
-												ALLOCSET_DEFAULT_SIZES);
+	if (state->sortopt & TUPLESORT_ALLOWBOUNDED)
+		state->tuplecontext = AllocSetContextCreate(state->sortcontext,
+													"Caller tuples",
+													ALLOCSET_DEFAULT_SIZES);
+	else
+		state->tuplecontext = GenerationContextCreate(state->sortcontext,
+													  "Caller tuples",
+													  ALLOCSET_DEFAULT_SIZES);
+
 
 	state->status = TSS_INITIAL;
 	state->bounded = false;
@@ -1444,6 +1454,8 @@ tuplesort_set_bound(Tuplesortstate *state, int64 bound)
 {
 	/* Assert we're called before loading any tuples */
 	Assert(state->status == TSS_INITIAL && state->memtupcount == 0);
+	/* Assert we allow bounded sorts */
+	Assert(state->sortopt & TUPLESORT_ALLOWBOUNDED);
 	/* Can't set the bound twice, either */
 	Assert(!state->bounded);
 	/* Also, this shouldn't be called in a parallel worker */
diff --git a/src/include/utils/tuplesort.h b/src/include/utils/tuplesort.h
index 345f4ce8024..364cf132fcb 100644
--- a/src/include/utils/tuplesort.h
+++ b/src/include/utils/tuplesort.h
@@ -92,6 +92,9 @@ typedef enum
 /* specifies whether non-sequential access to the sort result is required */
 #define	TUPLESORT_RANDOMACCESS			(1 << 0)
 
+/* specifies if the tuplesort is able to support bounded sorts */
+#define TUPLESORT_ALLOWBOUNDED			(1 << 1)
+
 typedef struct TuplesortInstrumentation
 {
 	TuplesortMethod sortMethod; /* sort algorithm used */