diff --git a/paddle/fluid/framework/ir/sync_batch_norm_pass.cc b/paddle/fluid/framework/ir/sync_batch_norm_pass.cc
index 90c1b23fe4f61eb8b6a70f7a6ab7de89754fb102..222c73761b4f5d960b1efafdb2e4368c17519fec 100644
--- a/paddle/fluid/framework/ir/sync_batch_norm_pass.cc
+++ b/paddle/fluid/framework/ir/sync_batch_norm_pass.cc
@@ -24,6 +24,10 @@ namespace ir {
 class SyncBatchNormPass : public Pass {
  protected:
   void ApplyImpl(ir::Graph *graph) const override {
+#if defined(_WIN32)
+    VLOG(3) << "Not use synchronize batch norm on windows";
+    return;
+#endif
     VLOG(3) << "Use synchronize batch norm";
     for (const Node *n : graph->Nodes()) {
       if (n->IsOp() && n->Op()) {
diff --git a/paddle/fluid/operators/CMakeLists.txt b/paddle/fluid/operators/CMakeLists.txt
index ea0c0b82be40ed488d7d649abb1706b3cbc794f2..f59e73d142a4ecea2d4e2208d649c3eeca852960 100644
--- a/paddle/fluid/operators/CMakeLists.txt
+++ b/paddle/fluid/operators/CMakeLists.txt
@@ -68,12 +68,8 @@ if (WITH_GPU)
     else()
         op_library(warpctc_op DEPS dynload_warpctc sequence_padding sequence_scale)
     endif()
-    if (NOT WIN32)
-        if (WITH_NCCL)
-            op_library(sync_batch_norm_op)
-            file(APPEND ${pybind_file} "USE_CUDA_ONLY_OP(sync_batch_norm);\n")
-        endif()
-    endif()
+    op_library(sync_batch_norm_op)
+    file(APPEND ${pybind_file} "USE_CUDA_ONLY_OP(sync_batch_norm);\n")
 else()
     op_library(warpctc_op DEPS dynload_warpctc sequence_padding sequence_scale)
 endif()