diff --git a/example/exampleapp/tests.py b/example/exampleapp/tests.py
index b66a8c9..0fd22c2 100644
--- a/example/exampleapp/tests.py
+++ b/example/exampleapp/tests.py
@@ -288,7 +288,7 @@ def test_sanity_checks(self):
         self.assertRaisesMessage(
             ValueError,
             'All fast_update() objects must have a primary key set.',
-            lambda : FieldUpdate.objects.fast_update([FieldUpdate(**EXAMPLE), FieldUpdate(**EXAMPLE)], ['f_char'])
+            lambda : FieldUpdate.objects.fast_update([FieldUpdate(**EXAMPLE)], ['f_char'])
         )
         # non concrete field
         mbase = MultiBase.objects.create()
@@ -305,37 +305,18 @@ def test_sanity_checks(self):
         )
 
 
-class TestFilterDuplicates(TestCase):
-    def test_apply_first_duplicate_only(self):
+class TestDuplicates(TestCase):
+    def test_raise_on_duplicates(self):
         a = FieldUpdate.objects.create()
-        updated = FieldUpdate.objects.fast_update([
-            FieldUpdate(pk=a.pk, **EXAMPLE),    # all values are trueish
-            FieldUpdate(pk=a.pk)                # all values None
-        ], FIELDS)
-        # only 1 row updated
-        self.assertEqual(updated, 1)
-        v = FieldUpdate.objects.all().values().first()
-        # all values should be trueish
-        self.assertEqual(all(e for e in v.values()), True)
+        with self.assertRaisesMessage(ValueError, 'cannot update duplicates'):
+            FieldUpdate.objects.fast_update([
+                FieldUpdate(pk=a.pk),
+                FieldUpdate(pk=a.pk)
+            ], FIELDS)
 
-    def test_multiple_duplicates(self):
-        a = FieldUpdate.objects.create()
-        b = FieldUpdate.objects.create()
-        c = FieldUpdate.objects.create()
-        updated = FieldUpdate.objects.fast_update([
-            FieldUpdate(pk=a.pk, **EXAMPLE),    # all values are trueish
-            FieldUpdate(pk=a.pk),               # all values None
-            FieldUpdate(pk=a.pk),
-            FieldUpdate(pk=b.pk, **EXAMPLE),    # all values are trueish
-            FieldUpdate(pk=a.pk),
-            FieldUpdate(pk=a.pk),
-            FieldUpdate(pk=b.pk),
-            FieldUpdate(pk=c.pk, **EXAMPLE)     # all values are trueish
-        ], FIELDS)
-        # 3 row updated
-        self.assertEqual(updated, 3)
-        v = list(FieldUpdate.objects.all().values())
-        # all values should be trueish
-        self.assertEqual(all(e for e in v[0].values()), True)
-        self.assertEqual(all(e for e in v[1].values()), True)
-        self.assertEqual(all(e for e in v[2].values()), True)
+    def test_no_pk_duplicates(self):
+        with self.assertRaisesMessage(ValueError, 'cannot update duplicates'):
+            FieldUpdate.objects.fast_update([
+                FieldUpdate(),
+                FieldUpdate()
+            ], FIELDS)
diff --git a/example/postgres_tests/tests.py b/example/postgres_tests/tests.py
index 63a2c4d..fa563a2 100644
--- a/example/postgres_tests/tests.py
+++ b/example/postgres_tests/tests.py
@@ -436,11 +436,21 @@ def test_updatefull_multiple(self):
         for r in results[1:]:
             for f in CU_FIELDS:
                 self.assertEqual(r[f], first[f])
-        # force threaded write
-        update_c = update_b * 100
-        FieldUpdateNotNull.objects.copy_update(update_c, CU_FIELDS)
+
+    def test_updatefull_multiple_threaded(self):
+        objs = []
+        for _ in range(10000):
+            objs.append(FieldUpdateNotNull())
+        FieldUpdateNotNull.objects.bulk_create(objs)
+        changeset = []
+        for o in objs:
+            changeset.append(FieldUpdateNotNull(pk=o.pk, **CU_EXAMPLE))
+
+        # force copy_update to use threaded logic due to payload >>64kB
+        FieldUpdateNotNull.objects.copy_update(changeset, CU_FIELDS)
         results = list(FieldUpdateNotNull.objects.all().values(*CU_FIELDS))
-        for r in results[201:]:
+        first = results[0]
+        for r in results[1:]:
             for f in CU_FIELDS:
                 self.assertEqual(r[f], first[f])
 
diff --git a/fast_update/fast.py b/fast_update/fast.py
index 7a1337d..ffd44a9 100644
--- a/fast_update/fast.py
+++ b/fast_update/fast.py
@@ -15,7 +15,7 @@
 """
 DB vendor low level interfaces
 
-To register a fast update implementations, call:
+To register fast update implementations, call:
 
     register_implementation('vendor', check_function)
 
@@ -70,44 +70,7 @@ def prepare_data_xy(
 """
 
 
-# memorize fast_update vendor on connection object
-SEEN_CONNECTIONS = cast(Dict[BaseDatabaseWrapper, str], WeakKeyDictionary())
-CHECKER = {}
-
-def register_implementation(
-    vendor: str,
-    func: Callable[[BaseDatabaseWrapper], Tuple[Any]]
-) -> None:
-    """
-    Register fast update implementation for db vendor.
-
-    `vendor` is the vendor name as returned by `connection.vendor`.
-    `func` is a lazy called function to check support for a certain
-    implementation at runtime for `connection`. The function should return
-    a tuple of (create_sql, prepare_data | None) for supported backends,
-    otherwise an empty tuple (needed to avoid re-eval).
-    """
-    CHECKER[vendor] = func
-
-
-def get_impl(conn: BaseDatabaseWrapper) -> str:
-    """
-    Try to get a fast update implementation for `conn`.
-    Calls once the the check function of `register_implementation` and
-    memorizes its result for `conn`.
-    Returns a tuple (create_sql, prepare_data | None) for supported backends,
-    otherwise an empty tuple.
-    """
-    impl = SEEN_CONNECTIONS.get(conn)
-    if impl is not None:
-        return impl
-    check = CHECKER.get(conn.vendor)
-    if not check:   # pragma: no cover
-        SEEN_CONNECTIONS[conn] = tuple()
-        return tuple()
-    impl = check(conn) or tuple()   # NOTE: in case check returns something nullish
-    SEEN_CONNECTIONS[conn] = impl
-    return impl
+# Fast update implementations for postgres, sqlite and mysql.
 
 
 def pq_cast(tname: str, field: Field, compiler: SQLCompiler, connection: Any) -> str:
@@ -243,14 +206,57 @@ def as_mysql(
     )
 
 
-# Register our default db implementations.
+# Implementation registry.
+
+
+# memorize fast_update vendor on connection object
+SEEN_CONNECTIONS = cast(Dict[BaseDatabaseWrapper, str], WeakKeyDictionary())
+CHECKER = {}
+
+def register_implementation(
+    vendor: str,
+    func: Callable[[BaseDatabaseWrapper], Tuple[Any]]
+) -> None:
+    """
+    Register fast update implementation for db vendor.
+
+    `vendor` is the vendor name as returned by `connection.vendor`.
+    `func` is a lazy called function to check support for a certain
+    implementation at runtime for `connection`. The function should return
+    a tuple of (create_sql, prepare_data | None) for supported backends,
+    otherwise an empty tuple (needed to avoid re-eval).
+    """
+    CHECKER[vendor] = func
+
+
+def get_impl(conn: BaseDatabaseWrapper) -> str:
+    """
+    Try to get a fast update implementation for `conn`.
+    Calls once the check function of `register_implementation` and
+    memorizes its result for `conn`.
+    Returns a tuple (create_sql, prepare_data | None) for supported backends,
+    otherwise an empty tuple.
+    """
+    impl = SEEN_CONNECTIONS.get(conn)
+    if impl is not None:
+        return impl
+    check = CHECKER.get(conn.vendor)
+    if not check:   # pragma: no cover
+        SEEN_CONNECTIONS[conn] = tuple()
+        return tuple()
+    impl = check(conn) or tuple()   # NOTE: in case check returns something nullish
+    SEEN_CONNECTIONS[conn] = impl
+    return impl
+
+
+# Register default db implementations from above.
 register_implementation(
     'postgresql',
     lambda _: (as_postgresql, None)
 )
 register_implementation(
     'sqlite',
-    # NOTE: check function does not handle versions <2.15 anymore
+    # NOTE: check function does not handle versions <3.15 anymore
     lambda conn: (as_sqlite, None) if conn.Database.sqlite_version_info >= (3, 33)
         else (as_sqlite_cte, prepare_data_sqlite_cte)
 )
@@ -260,6 +266,9 @@ def as_mysql(
 )
 
 
+# Update implementation.
+
+
 def update_from_values(
     c: CursorWrapper,
     tname: str,
@@ -344,14 +353,9 @@ def fast_update(
         with conn.cursor() as c:
             data = []
             counter = 0
-            seen = set()
             for o in objs:
-                row = [p(v, conn) for p, v in zip(prep_save, get(o))]
-                # filter for first batch occurence
-                if row[0] not in seen:
-                    counter += 1
-                    data += row
-                    seen.add(row[0])
+                counter += 1
+                data += [p(v, conn) for p, v in zip(prep_save, get(o))]
                 if counter >= batch_size_adjusted:
                     rows_updated += update_from_values(
                         c, model._meta.db_table, pk_field, fields,
@@ -359,7 +363,6 @@ def fast_update(
                     )
                     data = []
                     counter = 0
-                    seen = set()
             if data:
                 rows_updated += update_from_values(
                     c, model._meta.db_table, pk_field, fields,
diff --git a/fast_update/query.py b/fast_update/query.py
index 10b00fd..f315789 100644
--- a/fast_update/query.py
+++ b/fast_update/query.py
@@ -10,28 +10,33 @@ def sanity_check(
     model: Type[Model],
     objs: Iterable[Model],
     fields: Iterable[str],
+    op: str,
     batch_size: Optional[int] = None
 ) -> None:
     # basic sanity checks (most taken from bulk_update)
     if batch_size is not None and batch_size < 0:
         raise ValueError('Batch size must be a positive integer.')
     if not fields:
-        raise ValueError('Field names must be given to fast_update().')
-    if any(obj.pk is None for obj in objs):
-        raise ValueError('All fast_update() objects must have a primary key set.')
+        raise ValueError(f'Field names must be given to {op}.')
+    pks = set(obj.pk for obj in objs)
+    if len(pks) < len(objs):
+        raise ValueError(f'{op} cannot update duplicates.')
+    if None in pks:
+        raise ValueError(f'All {op} objects must have a primary key set.')
     fields_ = [model._meta.get_field(name) for name in fields]
     if any(not f.concrete or f.many_to_many for f in fields_):
-        raise ValueError('fast_update() can only be used with concrete fields.')
+        raise ValueError(f'{op} can only be used with concrete fields.')
     if any(f.primary_key for f in fields_):
-        raise ValueError('fast_update() cannot be used with primary key fields.')
+        raise ValueError(f'{op} cannot be used with primary key fields.')
     for obj in objs:
+        # TODO: This is really heavy in the runtime books, any elegant way to speedup?
         # TODO: django main has an additional argument 'fields' (saves some runtime?)
         obj._prepare_related_fields_for_save(operation_name='fast_update')
         # additionally raise on f-expression
         for field in fields_:
-            attr = getattr(obj, field.attname)
-            if hasattr(attr, 'resolve_expression'):
-                raise ValueError('fast_update() cannot be used with f-expressions.')
+            # TODO: use faster attrgetter
+            if hasattr(getattr(obj, field.attname), 'resolve_expression'):
+                raise ValueError(f'{op} cannot be used with f-expressions.')
 
 
 class FastUpdateQuerySet(QuerySet):
@@ -62,7 +67,7 @@ def fast_update(
             return 0
         objs = tuple(objs)
         fields_ = set(fields or [])
-        sanity_check(self.model, objs, fields_, batch_size)
+        sanity_check(self.model, objs, fields_, 'fast_update()', batch_size)
         return fast_update(self, objs, fields_, batch_size)
 
     fast_update.alters_data = True
@@ -111,7 +116,7 @@ def copy_update(
             return 0
         objs = tuple(objs)
         fields_ = set(fields or [])
-        sanity_check(self.model, objs, fields_)
+        sanity_check(self.model, objs, fields_, 'copy_update()')
         return copy_update(self, objs, fields_, field_encoders, encoding)
     
     copy_update.alters_data = True