diff --git a/src/realm/query.cpp b/src/realm/query.cpp
index d39bc855645..029010f3290 100644
--- a/src/realm/query.cpp
+++ b/src/realm/query.cpp
@@ -1398,9 +1398,12 @@ TableView Query::find_all(size_t limit) const
 
     TableView ret(*this, limit);
     if (m_ordering) {
+        // apply_descriptor_ordering will call do_sync
         ret.apply_descriptor_ordering(*m_ordering);
     }
-    ret.do_sync();
+    else {
+        ret.do_sync();
+    }
     return ret;
 }
 
diff --git a/src/realm/sort_descriptor.cpp b/src/realm/sort_descriptor.cpp
index 65ab201fa9a..f3caff82c45 100644
--- a/src/realm/sort_descriptor.cpp
+++ b/src/realm/sort_descriptor.cpp
@@ -301,7 +301,29 @@ BaseDescriptor::Sorter SortDescriptor::sorter(Table const& table, const IndexPai
 
 void SortDescriptor::execute(IndexPairs& v, const Sorter& predicate, const BaseDescriptor* next) const
 {
-    std::sort(v.begin(), v.end(), std::ref(predicate));
+    size_t limit = size_t(-1);
+    if (next && next->get_type() == DescriptorType::Limit) {
+        limit = static_cast<const LimitDescriptor*>(next)->get_limit();
+    }
+    // Measurements shows that if limit is smaller than size / 16, then
+    // it is quicker to make a sorted insert into a smaller vector
+    if (limit < (v.size() >> 4)) {
+        IndexPairs buffer;
+        buffer.reserve(limit + 1);
+        for (auto& elem : v) {
+            auto it = std::lower_bound(buffer.begin(), buffer.end(), elem, std::ref(predicate));
+            buffer.insert(it, elem);
+            if (buffer.size() > limit) {
+                buffer.pop_back();
+            }
+        }
+        v.m_removed_by_limit += v.size() - limit;
+        v.erase(v.begin() + limit, v.end());
+        std::move(buffer.begin(), buffer.end(), v.begin());
+    }
+    else {
+        std::sort(v.begin(), v.end(), std::ref(predicate));
+    }
 
     // not doing this on the last step is an optimisation
     if (next) {
diff --git a/test/benchmark-common-tasks/main.cpp b/test/benchmark-common-tasks/main.cpp
index 4f18f794acb..b3058f37433 100644
--- a/test/benchmark-common-tasks/main.cpp
+++ b/test/benchmark-common-tasks/main.cpp
@@ -1391,6 +1391,49 @@ struct BenchmarkSortIntDictionary : Benchmark {
     std::vector<size_t> m_indices;
 };
 
+struct BenchmarkSortThenLimit : Benchmark {
+    const char* name() const
+    {
+        return "SortThenLimit";
+    }
+
+    void before_all(DBRef group)
+    {
+        WriteTransaction tr(group);
+        TableRef t = tr.add_table(name());
+        m_col = t->add_column(type_Int, "first");
+
+        std::vector<int> values(10000);
+        std::iota(values.begin(), values.end(), 0);
+        std::shuffle(values.begin(), values.end(), std::mt19937(std::random_device()()));
+
+        for (auto i : values) {
+            t->create_object().set(m_col, i);
+        }
+        tr.commit();
+    }
+
+    void after_all(DBRef db)
+    {
+        WriteTransaction tr(db);
+        tr.get_group().remove_table(name());
+        tr.commit();
+    }
+
+    void operator()(DBRef db)
+    {
+        realm::ReadTransaction tr(db);
+        auto tv = tr.get_group().get_table(name())->where().find_all();
+        DescriptorOrdering ordering;
+        ordering.append_sort(SortDescriptor({{m_col}}));
+        ordering.append_limit(100);
+
+        tv.apply_descriptor_ordering(ordering);
+    }
+
+    ColKey m_col;
+};
+
 struct BenchmarkInsert : BenchmarkWithStringsTable {
     const char* name() const
     {
@@ -2249,7 +2292,6 @@ int benchmark_common_tasks_main()
 
 #define BENCH(B) run_benchmark<B>(results)
 #define BENCH2(B, mode) run_benchmark<B>(results, mode)
-
     BENCH2(BenchmarkEmptyCommit, true);
     BENCH2(BenchmarkEmptyCommit, false);
     BENCH2(BenchmarkNonInitiatorOpen, true);
@@ -2267,6 +2309,7 @@ int benchmark_common_tasks_main()
     BENCH(BenchmarkSortInt);
     BENCH(BenchmarkSortIntList);
     BENCH(BenchmarkSortIntDictionary);
+    BENCH(BenchmarkSortThenLimit);
 
     BENCH(BenchmarkUnorderedTableViewClear);
     BENCH(BenchmarkUnorderedTableViewClearIndexed);
diff --git a/test/test_table_view.cpp b/test/test_table_view.cpp
index c517c9ab9af..4459ede835f 100644
--- a/test/test_table_view.cpp
+++ b/test/test_table_view.cpp
@@ -24,6 +24,7 @@
 #include <sstream>
 #include <ostream>
 #include <cwchar>
+#include <chrono>
 
 #include <realm.hpp>
 
@@ -32,9 +33,13 @@
 #include "test.hpp"
 #include "test_table_helper.hpp"
 
+using namespace std::chrono;
+
 using namespace realm;
 using namespace test_util;
 
+extern unsigned int unit_test_random_seed;
+
 // Test independence and thread-safety
 // -----------------------------------
 //
@@ -2833,4 +2838,35 @@ TEST(TableView_CopyKeyValues)
     CHECK_EQUAL(yet_another_view.get_key(0), ObjKey(0));
 }
 
+TEST(TableView_SortFollowedByLimit)
+{
+    constexpr int limit = 100;
+    Table table;
+    auto col = table.add_column(type_Int, "first");
+    std::vector<int> values(10000);
+    std::iota(values.begin(), values.end(), 0);
+    std::shuffle(values.begin(), values.end(), std::mt19937(unit_test_random_seed));
+
+    for (auto i : values) {
+        table.create_object().set(col, i);
+    }
+
+    auto tv = table.where().find_all();
+    DescriptorOrdering ordering;
+    ordering.append_sort(SortDescriptor({{col}}));
+    ordering.append_limit(limit);
+
+    auto t1 = steady_clock::now();
+    tv.apply_descriptor_ordering(ordering);
+    auto t2 = steady_clock::now();
+
+    CHECK(t2 > t1);
+    // std::cout << duration_cast<microseconds>(t2 - t1).count() << " us" << std::endl;
+
+    CHECK_EQUAL(tv.size(), limit);
+    for (int i = 0; i < limit; i++) {
+        CHECK_EQUAL(tv.get_object(i).get<Int>(col), i);
+    }
+}
+
 #endif // TEST_TABLE_VIEW