diff --git a/paddle/phi/core/distributed/nccl_comm_context.cc b/paddle/phi/core/distributed/nccl_comm_context.cc
index faf29add30d916..bd49f0cff17086 100644
--- a/paddle/phi/core/distributed/nccl_comm_context.cc
+++ b/paddle/phi/core/distributed/nccl_comm_context.cc
@@ -231,6 +231,7 @@ void NCCLCommContext::GroupStart() {
 }
 void NCCLCommContext::GroupEnd() { NCCL_CHECK(phi::dynload::ncclGroupEnd()); }
 
+#if NCCL_VERSION_CODE >= 21100
 void NCCLCommContext::RedOpCreatePreMulSum(ncclRedOp_t* op,
                                            void* scalar,
                                            ncclDataType_t dtype,
@@ -242,6 +243,7 @@ void NCCLCommContext::RedOpCreatePreMulSum(ncclRedOp_t* op,
 void NCCLCommContext::RedOpDestroy(ncclRedOp_t op) {
   PADDLE_ENFORCE_GPU_SUCCESS(phi::dynload::ncclRedOpDestroy(op, nccl_comm_));
 }
+#endif
 
 }  // namespace distributed
 }  // namespace phi
diff --git a/paddle/phi/core/distributed/nccl_comm_context.h b/paddle/phi/core/distributed/nccl_comm_context.h
index 61c3fb06c0e33b..615112481ce2bd 100644
--- a/paddle/phi/core/distributed/nccl_comm_context.h
+++ b/paddle/phi/core/distributed/nccl_comm_context.h
@@ -98,16 +98,18 @@ class NCCLCommContext final : public CommContext {
               int root,
               gpuStream_t stream);
 
+  void GroupStart();
+
+  void GroupEnd();
+
+#if NCCL_VERSION_CODE >= 21100
   void RedOpCreatePreMulSum(ncclRedOp_t* op,
                             void* scalar,
                             ncclDataType_t dtype,
                             ncclScalarResidence_t residence);
 
   void RedOpDestroy(ncclRedOp_t op);
-
-  void GroupStart();
-
-  void GroupEnd();
+#endif
 
  private:
   DISABLE_COPY_AND_ASSIGN(NCCLCommContext);
diff --git a/test/legacy_test/distributed_fused_lamb_test_base.py b/test/legacy_test/distributed_fused_lamb_test_base.py
index baffc7dd5e5460..ec45942bd0aa16 100644
--- a/test/legacy_test/distributed_fused_lamb_test_base.py
+++ b/test/legacy_test/distributed_fused_lamb_test_base.py
@@ -270,7 +270,10 @@ def setUpClass(cls):
         paddle.enable_static()
         paddle.set_flags({'FLAGS_cudnn_deterministic': True})
         _clip_by_global_norm_using_mp_type(True)
-        fleet.init(role_maker=get_role_maker())
+        if os.environ.get("FLAGS_dynamic_static_unified_comm") == "1":
+            fleet.init(role_maker=get_role_maker())
+        else:
+            paddle.distributed.collective._init_parallel_env("nccl")
 
     def config(self):
         clip_after_allreduce = bool(