diff --git a/test/npu/test_npu.py b/test/npu/test_npu.py
index 67834e9cf64a76e7cf35d11151ad0d17e10cd970..f53664a5fb554d1dc9bb96cac2a91324ab8d3676 100644
--- a/test/npu/test_npu.py
+++ b/test/npu/test_npu.py
@@ -449,6 +449,7 @@ class TestNpu(TestCase):
         stream.record_event(start_event)
         stream.record_event(event)
         event.synchronize()
+        #just test
         self.assertTrue(event.query())
         self.assertGreater(start_event.elapsed_time(event), 0)
 
diff --git a/torch_npu/csrc/framework/OpParamMaker.cpp b/torch_npu/csrc/framework/OpParamMaker.cpp
index ef7da6d204faf1b4c8495330fcb8878562e18721..bf552e138a4e14e13b5869398107d5bb0f7dd1d1 100644
--- a/torch_npu/csrc/framework/OpParamMaker.cpp
+++ b/torch_npu/csrc/framework/OpParamMaker.cpp
@@ -450,8 +450,12 @@ int MemcopyAsyncFunc(c10_npu::queue::QueueParas *in, aclrtStream stream)
 {
     auto cur_paras = static_cast<c10_npu::queue::CopyParas *>(in->paramVal);
     logger->debug("MemcopyAsyncFunc Run.");
-    aclError ret =
-        aclrtMemcpyAsync(cur_paras->dst, cur_paras->dstLen, cur_paras->src, cur_paras->srcLen, cur_paras->kind, stream);
+    aclError ret;
+    if (c10_npu::acl::AclrtMemcpyAsyncWithConditionExist() && cur_paras->kind == aclrtMemcpyKind::ACL_MEMCPY_DEVICE_TO_HOST) {
+        ret = aclrtMemcpyAsyncWithCondition(cur_paras->dst, cur_paras->dstLen, cur_paras->src, cur_paras->srcLen, cur_paras->kind, stream);
+    } else {
+        ret = aclrtMemcpyAsync(cur_paras->dst, cur_paras->dstLen, cur_paras->src, cur_paras->srcLen, cur_paras->kind, stream);
+    }
     if (ret != ACL_ERROR_NONE) {
         auto ret_temp = c10_npu::acl::AclrtPeekAtLastError(ACL_RT_THREAD_LEVEL);
         if (ret_temp != ACL_ERROR_NONE) {