diff --git a/documentation/release_6.4.htm b/documentation/release_6.4.htm
index dd3647470..2a2b979cb 100644
--- a/documentation/release_6.4.htm
+++ b/documentation/release_6.4.htm
@@ -65,7 +65,7 @@ <h3>Changed functionality</h3>
       <ul>
         <li>
           Compatibility with <a href="https://github.com/KUL-recon-lab/libparallelproj/">libparallelproj 2.0</a> and usage of CUDA managed pointers via
-          <a href="https://github.com/AMYPAD/cuvec">CuVec</a> for internal variables in our parallelproj interface and <code>CudaGibbsPenalty</code>.
+          <a href="https://github.com/AMYPAD/CuVec">CuVec</a> for internal variables in our parallelproj interface and <code>CudaGibbsPenalty</code>.
           This results in a ~20% speed-up, but also code simplification.
           <br>
           <a href=https://github.com/UCL/STIR/pull/1689>PR #1689</a>.
diff --git a/src/include/stir/cuda_utilities.h b/src/include/stir/cuda_utilities.h
index be7f11e88..d5ec02727 100644
--- a/src/include/stir/cuda_utilities.h
+++ b/src/include/stir/cuda_utilities.h
@@ -87,6 +87,7 @@ template <int num_dimensions, typename elemT>
 inline void
 array_to_device(CuVec<elemT>& dev_data, const Array<num_dimensions, elemT>& stir_array)
 {
+  dev_data.resize(stir_array.size_all());
   std::copy(stir_array.begin_all(), stir_array.end_all(), dev_data.begin());
 }
 
@@ -127,6 +128,8 @@ array_to_host(Array<num_dimensions, elemT>& stir_array, const CuVec<elemT>& dev_
 {
   if (sync)
     cudaDeviceSynchronize();
+  if (stir_array.size_all() != dev_data.size())
+    error("array_to_host: size mismatch between CuVec and Array");
   std::copy(dev_data.begin(), dev_data.end(), stir_array.begin_all());
 }