diff --git a/Dockerfile b/Dockerfile
index 136db772cc6a24b8084120fa6bab666bc1eda78e..150344a8116e2be9b5bab8e5fdcc9c37f4025020 100644
--- a/Dockerfile
+++ b/Dockerfile
@@ -22,7 +22,7 @@ COPY ./paddle/scripts/docker/root/ /root/
 
 RUN apt-get update && \
     apt-get install -y \
-    git python-pip python-dev openssh-server bison  \
+    git python-pip python-dev openssh-server bison libnccl-dev \
     wget unzip unrar tar xz-utils bzip2 gzip coreutils ntp \
     curl sed grep graphviz libjpeg-dev zlib1g-dev  \
     python-matplotlib gcc-4.8 g++-4.8 \
diff --git a/paddle/framework/CMakeLists.txt b/paddle/framework/CMakeLists.txt
index 774c7b021754be607cd895ca910583e992ed26a0..dbe76a8eaf134f7db08fb545297c8e4db68a7aab 100644
--- a/paddle/framework/CMakeLists.txt
+++ b/paddle/framework/CMakeLists.txt
@@ -19,15 +19,15 @@ cc_test(scope_test SRCS scope_test.cc DEPS scope)
 proto_library(framework_proto SRCS framework.proto)
 
 cc_library(attribute SRCS attribute.cc DEPS framework_proto)
-cc_library(proto_desc SRCS var_desc.cc op_desc.cc block_desc.cc program_desc.cc DEPS attribute ddim op_info)
 cc_test(program_desc_test SRCS program_desc_test.cc DEPS proto_desc)
 cc_library(op_proto_maker SRCS op_proto_maker.cc DEPS framework_proto attribute)
 cc_test(op_proto_maker_test SRCS op_proto_maker_test.cc DEPS op_proto_maker)
 cc_library(op_info SRCS op_info.cc DEPS attribute framework_proto)
-cc_library(operator SRCS operator.cc DEPS op_info device_context tensor scope proto_desc glog)
+cc_library(operator SRCS operator.cc DEPS op_info device_context tensor scope glog)
 cc_test(operator_test SRCS operator_test.cc DEPS operator op_registry)
+cc_library(proto_desc SRCS var_desc.cc op_desc.cc block_desc.cc program_desc.cc DEPS attribute ddim op_info operator)
 
-cc_library(op_registry SRCS op_registry.cc DEPS op_proto_maker op_info operator glog)
+cc_library(op_registry SRCS op_registry.cc DEPS op_proto_maker op_info operator glog proto_desc)
 cc_test(op_registry_test SRCS op_registry_test.cc DEPS op_registry)
 
 py_proto_compile(framework_py_proto SRCS framework.proto)
diff --git a/paddle/framework/operator.cc b/paddle/framework/operator.cc
index 2fca816f353635d3bff184323755961ee82fbb68..a67625fa88fd2fbe4db43241ee824519ceac7017 100644
--- a/paddle/framework/operator.cc
+++ b/paddle/framework/operator.cc
@@ -252,5 +252,20 @@ std::ostream& operator<<(std::ostream& os,
   return os;
 }
 
+bool OpSupportGPU(const std::string& op_type) {
+  auto& all_kernels = OperatorWithKernel::AllOpKernels();
+  auto it = all_kernels.find(op_type);
+  if (it == all_kernels.end()) {
+    // All control operator must support GPU
+    return true;
+  }
+  for (auto& kern_pair : it->second) {
+    if (platform::is_gpu_place(kern_pair.first.place_)) {
+      return true;
+    }
+  }
+  return false;
+}
+
 }  // namespace framework
 }  // namespace paddle
diff --git a/paddle/framework/operator.h b/paddle/framework/operator.h
index 12cd307297d010201a47e048089ed7be0db52647..9d7fe1f5ba293227e67cf6bfcd566a1247c567ed 100644
--- a/paddle/framework/operator.h
+++ b/paddle/framework/operator.h
@@ -649,5 +649,7 @@ class OperatorWithKernel : public OperatorBase {
 std::ostream& operator<<(std::ostream& os,
                          const OperatorWithKernel::OpKernelKey& kernel_key);
 
+extern bool OpSupportGPU(const std::string& op_type);
+
 }  // namespace framework
 }  // namespace paddle
diff --git a/paddle/pybind/pybind.cc b/paddle/pybind/pybind.cc
index e5ddc14587623905dbf52b4c1690236ffeb069a1..26b793a4bbf5df7a2635838a6c6a8264ca8ebb67 100644
--- a/paddle/pybind/pybind.cc
+++ b/paddle/pybind/pybind.cc
@@ -466,6 +466,8 @@ All parameter, weight, gradient are variables in Paddle.
   BindVarDsec(m);
   BindOpDesc(m);
 
+  m.def("op_support_gpu", OpSupportGPU);
+
   return m.ptr();
 }
 }  // namespace pybind
diff --git a/paddle/scripts/docker/build.sh b/paddle/scripts/docker/build.sh
index 2ac455d771bf78377ce4ee7d921393d3b3958e3c..a08716c5a559def54bb7b989f250b489f6a805a2 100644
--- a/paddle/scripts/docker/build.sh
+++ b/paddle/scripts/docker/build.sh
@@ -141,10 +141,17 @@ RUN sed -i '${APT_MIRROR}' /etc/apt/sources.list
 EOF
 fi
 
+if [[ ${WITH_GPU} == "ON"  ]]; then
+  NCCL_DEPS="apt-get install -y libnccl-dev &&"
+else
+  NCCL_DEPS="" 
+fi
+
 cat >> /paddle/build/Dockerfile <<EOF
 ADD python/dist/*.whl /
 # run paddle version to install python packages first
 RUN apt-get update &&\
+    ${NCCL_DEPS}\
     apt-get install -y wget python-pip && pip install -U pip && \
     pip install /*.whl; apt-get install -f -y && \
     apt-get clean -y && \
diff --git a/python/paddle/v2/framework/tests/test_op_support_gpu.py b/python/paddle/v2/framework/tests/test_op_support_gpu.py
new file mode 100644
index 0000000000000000000000000000000000000000..dd36c666c440a5c378dfceac4502cd8277417412
--- /dev/null
+++ b/python/paddle/v2/framework/tests/test_op_support_gpu.py
@@ -0,0 +1,11 @@
+import unittest
+import paddle.v2.framework.core as core
+
+
+class TestOpSupportGPU(unittest.TestCase):
+    def test_case(self):
+        self.assertEqual(core.is_compile_gpu(), core.op_support_gpu("sum"))
+
+
+if __name__ == '__main__':
+    unittest.main()