diff --git a/deploy/cpp/src/paddlex.cpp b/deploy/cpp/src/paddlex.cpp
index 5d33ae0e60285f41a2c24cca0ce96f51b54478bf..6d3c23094c3944b9359c701c4c3359c26313d1e3 100644
--- a/deploy/cpp/src/paddlex.cpp
+++ b/deploy/cpp/src/paddlex.cpp
@@ -66,7 +66,7 @@ void Model::create_predictor(const std::string& model_dir,
   if (key == "") {
     config.SetModel(model_file, params_file);
   }
-  if (use_mkl) {
+  if (use_mkl && !use_gpu) {
     if (name != "HRNet" && name != "DeepLabv3p" && name != "PPYOLO") {
         config.EnableMKLDNN();
         config.SetCpuMathLibraryNumThreads(mkl_thread_num);
diff --git a/paddlex/deploy.py b/paddlex/deploy.py
index 12570bec5c90b6334b41c77dac99e4b93fb144e4..b04f46ebf037af0d89e46f3d8e24efdb36a2fb47 100644
--- a/paddlex/deploy.py
+++ b/paddlex/deploy.py
@@ -108,7 +108,7 @@ class Predictor:
             config.enable_use_gpu(100, gpu_id)
         else:
             config.disable_gpu()
-        if use_mkl:
+        if use_mkl and not use_gpu:
             if self.model_name not in ["HRNet", "DeepLabv3p", "PPYOLO"]:
                 config.enable_mkldnn()
                 config.set_cpu_math_library_num_threads(mkl_thread_num)