diff --git a/cmake/external/lite.cmake b/cmake/external/lite.cmake
index 0031757467f3784c5fe72182f7aaf06b90acd6c2..206cbf1d8f8d18d30d4008f13b60e0ee5d1ad85a 100644
--- a/cmake/external/lite.cmake
+++ b/cmake/external/lite.cmake
@@ -106,7 +106,7 @@ if (NOT LITE_SOURCE_DIR OR NOT LITE_BINARY_DIR)
   else()
     set(LITE_BUILD_COMMAND $(MAKE) publish_inference -j)
     set(LITE_OPTIONAL_ARGS -DWITH_MKL=ON
-                           -DLITE_WITH_CUDA=${WITH_GPU}
+                           -DLITE_WITH_CUDA=OFF
                            -DWITH_MKLDNN=OFF
                            -DLITE_WITH_X86=ON
                            -DLITE_WITH_PROFILE=OFF
@@ -115,9 +115,6 @@ if (NOT LITE_SOURCE_DIR OR NOT LITE_BINARY_DIR)
                            -DWITH_PYTHON=OFF
                            -DWITH_TESTING=OFF
                            -DLITE_BUILD_EXTRA=ON
-                           -DCUDNN_ROOT=${CUDNN_ROOT}
-                           -DLITE_WITH_STATIC_CUDA=OFF
-                           -DCUDA_ARCH_NAME=${CUDA_ARCH_NAME}
                            -DLITE_WITH_XPU=${LITE_WITH_XPU}
                            -DXPU_SDK_URL=${XPU_BASE_URL}
                            -DXPU_SDK_ENV=${XPU_SDK_ENV}
diff --git a/paddle/fluid/inference/lite/engine.cc b/paddle/fluid/inference/lite/engine.cc
index cd78cfecd863577e61bd90e70cecfcc3e562f1da..c43e5c39bfeeb59947506b98cb2abb83c45e6ea8 100644
--- a/paddle/fluid/inference/lite/engine.cc
+++ b/paddle/fluid/inference/lite/engine.cc
@@ -12,10 +12,6 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-#if defined(PADDLE_WITH_CUDA) || defined(PADDLE_WITH_HIP)
-#define LITE_WITH_CUDA 1
-#endif
-
 #ifdef LITE_SUBGRAPH_WITH_XPU
 #define LITE_WITH_XPU 1
 #endif
diff --git a/paddle/fluid/inference/lite/test_tensor_utils.cc b/paddle/fluid/inference/lite/test_tensor_utils.cc
index b0c7c7448a50ef11b4c457c8f4c55196ceb0d453..4e240ef9e40b3f72232dbfa9ca019c8e16e4bd62 100644
--- a/paddle/fluid/inference/lite/test_tensor_utils.cc
+++ b/paddle/fluid/inference/lite/test_tensor_utils.cc
@@ -151,22 +151,12 @@ TEST(LiteEngineOp, TensorCopyAsync) {
   auto* ctx_cpu =
       platform::DeviceContextPool::Instance().Get(platform::CPUPlace());
   test_tensor_copy(*ctx_cpu);
-#if defined(PADDLE_WITH_CUDA) || defined(PADDLE_WITH_HIP)
-  auto* ctx_gpu =
-      platform::DeviceContextPool::Instance().Get(platform::CUDAPlace(0));
-  test_tensor_copy(*ctx_gpu);
-#endif
 }
 
 TEST(LiteEngineOp, TensorShare) {
   auto* ctx_cpu =
       platform::DeviceContextPool::Instance().Get(platform::CPUPlace());
   test_tensor_share(*ctx_cpu);
-#if defined(PADDLE_WITH_CUDA) || defined(PADDLE_WITH_HIP)
-  auto* ctx_gpu =
-      platform::DeviceContextPool::Instance().Get(platform::CUDAPlace(0));
-  test_tensor_share(*ctx_gpu);
-#endif
 }
 
 }  // namespace utils
diff --git a/paddle/fluid/inference/tests/api/lite_mul_model_test.cc b/paddle/fluid/inference/tests/api/lite_mul_model_test.cc
index 9211ea246a5c5e0cdc75e6fef72ae0e4e40d69af..6aaad4d4545d694077e676277e022afa0bae5ccb 100644
--- a/paddle/fluid/inference/tests/api/lite_mul_model_test.cc
+++ b/paddle/fluid/inference/tests/api/lite_mul_model_test.cc
@@ -119,35 +119,13 @@ TEST(AnalysisPredictor, lite_xpu) {
 }
 #endif
 
-#if defined(PADDLE_WITH_CUDA) || defined(PADDLE_WITH_HIP)
-TEST(AnalysisPredictor, thread_local_stream) {
-  const size_t thread_num = 5;
-  std::vector<std::thread> threads(thread_num);
-  Barrier barrier(thread_num);
-  for (size_t i = 0; i < threads.size(); ++i) {
-    threads[i] = std::thread([&barrier, i]() {
-      AnalysisConfig config;
-      config.EnableUseGpu(100, 0);
-      config.SetModel(FLAGS_infer_model + "/" + "mul_model");
-      config.EnableGpuMultiStream();
-      test_predictor(config, &barrier);
-      test_predictor_zero_copy(config);
-    });
-  }
-  for (auto& th : threads) {
-    th.join();
-  }
-}
-
 TEST(AnalysisPredictor, lite_engine) {
   AnalysisConfig config;
-  config.EnableUseGpu(100, 0);
   config.SetModel(FLAGS_infer_model + "/" + "mul_model");
   config.EnableLiteEngine(paddle::AnalysisConfig::Precision::kFloat32);
   test_predictor(config);
   test_predictor_zero_copy(config);
 }
-#endif
 
 }  // namespace inference
 }  // namespace paddle
diff --git a/paddle/fluid/inference/tests/api/lite_resnet50_test.cc b/paddle/fluid/inference/tests/api/lite_resnet50_test.cc
index 59bbaa2b78fb00428aa61bde3c65998721e757f3..e585257e485626e3b149e5b1341e19cb5c098e25 100644
--- a/paddle/fluid/inference/tests/api/lite_resnet50_test.cc
+++ b/paddle/fluid/inference/tests/api/lite_resnet50_test.cc
@@ -22,10 +22,9 @@ limitations under the License. */
 namespace paddle {
 namespace inference {
 
-TEST(AnalysisPredictor, use_gpu) {
+TEST(AnalysisPredictor, use_cpu) {
   std::string model_dir = FLAGS_infer_model + "/" + "model";
   AnalysisConfig config;
-  config.EnableUseGpu(100, 0);
   config.SetModel(model_dir + "/model", model_dir + "/params");
   config.EnableLiteEngine(paddle::AnalysisConfig::Precision::kFloat32, true);
 
@@ -73,10 +72,9 @@ TEST(AnalysisPredictor, use_gpu) {
 
 namespace paddle_infer {
 
-TEST(Predictor, use_gpu) {
+TEST(Predictor, use_cpu) {
   std::string model_dir = FLAGS_infer_model + "/" + "model";
   Config config;
-  config.EnableUseGpu(100, 0);
   config.SetModel(model_dir + "/model", model_dir + "/params");
   config.EnableLiteEngine(PrecisionType::kFloat32);