diff --git a/cuFFT/1d_mgpu_c2c/1d_mgpu_c2c_example.cpp b/cuFFT/1d_mgpu_c2c/1d_mgpu_c2c_example.cpp
index 52929e48..f07f27b3 100644
--- a/cuFFT/1d_mgpu_c2c/1d_mgpu_c2c_example.cpp
+++ b/cuFFT/1d_mgpu_c2c/1d_mgpu_c2c_example.cpp
@@ -105,6 +105,11 @@ void spmg(dim_t fft, int &batch_size, gpus_t gpus, cpudata_t &h_data_in, cpudata
     // Execute the plan
     CUFFT_CALL(cufftXtExecDescriptor(plan, indesc, indesc, CUFFT_FORWARD));
 
+#if CUFFT_VERSION >= 10400
+    // The transform runs on the plan's stream; wait for it before reading back
+    CUDA_RT_CALL(cudaStreamSynchronize(stream));
+#endif
+
     // Copy output data to CPU
     CUFFT_CALL(cufftXtMemcpy(plan, reinterpret_cast<void *>(h_data_out.data()),
                              reinterpret_cast<void *>(indesc), CUFFT_COPY_DEVICE_TO_HOST));
diff --git a/cuFFT/3d_mgpu_c2c/3d_mgpu_c2c_example.cpp b/cuFFT/3d_mgpu_c2c/3d_mgpu_c2c_example.cpp
index 55710490..517a922d 100644
--- a/cuFFT/3d_mgpu_c2c/3d_mgpu_c2c_example.cpp
+++ b/cuFFT/3d_mgpu_c2c/3d_mgpu_c2c_example.cpp
@@ -105,6 +105,11 @@ void spmg(dim_t fft, gpus_t gpus, cpudata_t &h_data_in, cpudata_t &h_data_out,
     // Execute the plan
     CUFFT_CALL(cufftXtExecDescriptor(plan, indesc, indesc, CUFFT_FORWARD));
 
+#if CUFFT_VERSION >= 10400
+    // The transform runs on the plan's stream; wait for it before reading back
+    CUDA_RT_CALL(cudaStreamSynchronize(stream));
+#endif
+
     // Copy output data to CPU
     CUFFT_CALL(cufftXtMemcpy(plan, reinterpret_cast<void *>(h_data_out.data()),
                              reinterpret_cast<void *>(indesc), CUFFT_COPY_DEVICE_TO_HOST));
diff --git a/cuFFT/3d_mgpu_r2c_c2r/3d_mgpu_r2c_c2r_example.cpp b/cuFFT/3d_mgpu_r2c_c2r/3d_mgpu_r2c_c2r_example.cpp
index 2cf5180a..8bf51f83 100644
--- a/cuFFT/3d_mgpu_r2c_c2r/3d_mgpu_r2c_c2r_example.cpp
+++ b/cuFFT/3d_mgpu_r2c_c2r/3d_mgpu_r2c_c2r_example.cpp
@@ -155,6 +155,11 @@ void spmg(dim_t fft, gpus_t gpus, cpudata_t &h_data_in, cpudata_t &h_data_out,
     // Execute the plan_r2c
     CUFFT_CALL(cufftXtExecDescriptor(plan_r2c, indesc, indesc, CUFFT_FORWARD));
 
+#if CUFFT_VERSION >= 10400
+    // The transform runs on the plan's stream; wait for it before reading back
+    CUDA_RT_CALL(cudaStreamSynchronize(stream));
+#endif
+
     // Scale complex results
     float scale{2.f};
     scaleComplex(indesc, scale, h_data_out.size(), gpus.size());
@@ -162,6 +167,11 @@ void spmg(dim_t fft, gpus_t gpus, cpudata_t &h_data_in, cpudata_t &h_data_out,
     // Execute the plan_c2r
     CUFFT_CALL(cufftXtExecDescriptor(plan_c2r, indesc, indesc, CUFFT_INVERSE));
 
+#if CUFFT_VERSION >= 10400
+    // The transform runs on the plan's stream; wait for it before reading back
+    CUDA_RT_CALL(cudaStreamSynchronize(stream));
+#endif
+
     // Copy output data to CPU
     CUFFT_CALL(cufftXtMemcpy(plan_c2r, (void *)h_data_out.data(), (void *)indesc,
                              CUFFT_COPY_DEVICE_TO_HOST));