diff --git a/unsupported/Eigen/CXX11/src/Tensor/TensorConversion.h b/unsupported/Eigen/CXX11/src/Tensor/TensorConversion.h
index e254c0b7b..21bb91d69 100644
--- a/unsupported/Eigen/CXX11/src/Tensor/TensorConversion.h
+++ b/unsupported/Eigen/CXX11/src/Tensor/TensorConversion.h
@@ -195,8 +195,11 @@ struct TensorEvaluator<const TensorConversionOp<TargetType, ArgType>, Device>
 
   EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Dimensions& dimensions() const { return m_impl.dimensions(); }
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE bool evalSubExprsIfNeeded(Scalar* /*data*/)
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE bool evalSubExprsIfNeeded(Scalar* data)
   {
+    if (internal::is_same<TargetType, SrcType>::value) {
+      return m_impl.evalSubExprsIfNeeded((SrcType*)data);
+    }
     m_impl.evalSubExprsIfNeeded(NULL);
     return true;
   }
diff --git a/unsupported/test/cxx11_tensor_of_float16_cuda.cu b/unsupported/test/cxx11_tensor_of_float16_cuda.cu
index ff045db7f..9b9fd843c 100644
--- a/unsupported/test/cxx11_tensor_of_float16_cuda.cu
+++ b/unsupported/test/cxx11_tensor_of_float16_cuda.cu
@@ -239,6 +239,8 @@ void test_cxx11_tensor_of_float16_cuda()
   Eigen::GpuDevice device(&stream);
   if (device.majorDeviceVersion() > 5 ||
       (device.majorDeviceVersion() == 5 && device.minorDeviceVersion() >= 3)) {
+    std::cout << "Running test on device with capability " << device.majorDeviceVersion() << "." << device.minorDeviceVersion() << std::endl;
+
     CALL_SUBTEST_1(test_cuda_conversion());
     CALL_SUBTEST_1(test_cuda_unary());
     CALL_SUBTEST_1(test_cuda_elementwise());