diff --git a/Eigen/src/Core/util/Macros.h b/Eigen/src/Core/util/Macros.h
index eebfd901d..961097baf 100644
--- a/Eigen/src/Core/util/Macros.h
+++ b/Eigen/src/Core/util/Macros.h
@@ -1198,7 +1198,7 @@ namespace Eigen {
   #define EIGEN_USING_STD(FUNC) using std::FUNC;
 #endif
 
-#if EIGEN_COMP_MSVC_STRICT && (EIGEN_COMP_MSVC < 1900 || (EIGEN_COMP_MSVC == 1900 && EIGEN_COMP_NVCC))
+#if EIGEN_COMP_MSVC_STRICT && (EIGEN_COMP_MSVC < 1916 || (EIGEN_COMP_MSVC == 1916 && EIGEN_COMP_NVCC))
   // For older MSVC versions, as well as 1900 && CUDA 8, using the base operator is necessary,
   //   otherwise we get duplicate definition errors
   // For later MSVC versions, we require explicit operator= definition, otherwise we get
diff --git a/unsupported/test/cxx11_tensor_of_float16_gpu.cu b/unsupported/test/cxx11_tensor_of_float16_gpu.cu
index 30bcc1d28..e11782a79 100644
--- a/unsupported/test/cxx11_tensor_of_float16_gpu.cu
+++ b/unsupported/test/cxx11_tensor_of_float16_gpu.cu
@@ -113,7 +113,7 @@ void test_gpu_unary() {
 
   gpu_float.device(gpu_device) = gpu_float.random() - gpu_float.constant(0.5f);
   gpu_res_float.device(gpu_device) = gpu_float.abs();
-  gpu_res_half.device(gpu_device) = gpu_float.cast<Eigen::half>().abs().cast<float>();
+  gpu_res_half.device(gpu_device) = gpu_float.cast<Eigen::half>().abs().template cast<float>();
 
   Tensor<float, 1> half_prec(num_elem);
   Tensor<float, 1> full_prec(num_elem);
@@ -154,7 +154,7 @@ void test_gpu_elementwise() {
   gpu_float1.device(gpu_device) = gpu_float1.random();
   gpu_float2.device(gpu_device) = gpu_float2.random();
   gpu_res_float.device(gpu_device) = (gpu_float1 + gpu_float2) * gpu_float1;
-  gpu_res_half.device(gpu_device) = ((gpu_float1.cast<Eigen::half>() + gpu_float2.cast<Eigen::half>()) * gpu_float1.cast<Eigen::half>()).cast<float>();
+  gpu_res_half.device(gpu_device) = ((gpu_float1.cast<Eigen::half>() + gpu_float2.cast<Eigen::half>()) * gpu_float1.cast<Eigen::half>()).template cast<float>();
 
   Tensor<float, 1> half_prec(num_elem);
   Tensor<float, 1> full_prec(num_elem);
@@ -444,8 +444,8 @@ void test_gpu_forced_evals() {
 
   gpu_float.device(gpu_device) = gpu_float.random() - gpu_float.constant(0.5f);
   gpu_res_float.device(gpu_device) = gpu_float.abs();
-  gpu_res_half1.device(gpu_device) = gpu_float.cast<Eigen::half>().abs().eval().cast<float>();
-  gpu_res_half2.device(gpu_device) = gpu_float.cast<Eigen::half>().abs().broadcast(no_bcast).eval().cast<float>();
+  gpu_res_half1.device(gpu_device) = gpu_float.cast<Eigen::half>().abs().eval().template cast<float>();
+  gpu_res_half2.device(gpu_device) = gpu_float.cast<Eigen::half>().abs().broadcast(no_bcast).eval().template cast<float>();
 
   Tensor<float, 1> half_prec1(num_elem);
   Tensor<float, 1> half_prec2(num_elem);