diff --git a/paddle/fluid/framework/details/nccl_all_reduce_op_handle.cc b/paddle/fluid/framework/details/nccl_all_reduce_op_handle.cc
index f2303ff4cabf31eb2fcf53bd24a2dfb8a7ab30a6..116b13d3301d780875580867a8d53e91b2781145 100644
--- a/paddle/fluid/framework/details/nccl_all_reduce_op_handle.cc
+++ b/paddle/fluid/framework/details/nccl_all_reduce_op_handle.cc
@@ -50,10 +50,6 @@ void NCCLAllReduceOpHandle::RunImpl() {
 
       auto &lod_tensor = s->FindVar(var_name)->Get<LoDTensor>();
       void *buffer = const_cast<void *>(lod_tensor.data<void>());
-      uintptr_t buf = reinterpret_cast<uintptr_t>(buffer);
-      if (buf % sizeof(float) != 0) {
-        VLOG(3) << "Buffer is not aligned " << buf;
-      }
 
       if (dtype == -1) {
         dtype = platform::ToNCCLDataType(lod_tensor.type());
diff --git a/paddle/fluid/platform/nccl_helper.h b/paddle/fluid/platform/nccl_helper.h
index 29990043206509e4192bfff84832f09ef127d9dd..ecdd98987dd41985c7ac77a7fceb356e15f6cd3b 100644
--- a/paddle/fluid/platform/nccl_helper.h
+++ b/paddle/fluid/platform/nccl_helper.h
@@ -36,12 +36,10 @@ inline ncclDataType_t ToNCCLDataType(std::type_index type) {
 
 class NCCLGroupGuard {
  public:
-  inline NCCLGroupGuard() {
-    mutex().lock();
-    PADDLE_ENFORCE(dynload::ncclGroupStart());
-  }
+  inline NCCLGroupGuard() { PADDLE_ENFORCE(dynload::ncclGroupStart()); }
 
   inline ~NCCLGroupGuard() {
+    mutex().lock();
     PADDLE_ENFORCE(dynload::ncclGroupEnd());
     mutex().unlock();
   }