diff --git a/cpp/src/strings/case.cu b/cpp/src/strings/case.cu
index c1688d20791..27befdea209 100644
--- a/cpp/src/strings/case.cu
+++ b/cpp/src/strings/case.cu
@@ -294,7 +294,7 @@ CUDF_KERNEL void has_multibytes_kernel(char const* d_input_chars,
   auto const mb_total = block_reduce(temp_storage).Reduce(mb_count, cub::Sum());
 
   if ((lane_idx == 0) && (mb_total > 0)) {
-    cuda::atomic_ref<int64_t, cuda::thread_scope_block> ref{*d_output};
+    cuda::atomic_ref<int64_t, cuda::thread_scope_device> ref{*d_output};
     ref.fetch_add(mb_total, cuda::std::memory_order_relaxed);
   }
 }
diff --git a/cpp/src/strings/replace/multi.cu b/cpp/src/strings/replace/multi.cu
index 8e5c5cf60b8..43a3d69091a 100644
--- a/cpp/src/strings/replace/multi.cu
+++ b/cpp/src/strings/replace/multi.cu
@@ -283,7 +283,7 @@ CUDF_KERNEL void count_targets(replace_multi_parallel_fn fn, int64_t chars_bytes
   auto const total = block_reduce(temp_storage).Reduce(count, cub::Sum());
 
   if ((lane_idx == 0) && (total > 0)) {
-    cuda::atomic_ref<int64_t, cuda::thread_scope_block> ref{*d_output};
+    cuda::atomic_ref<int64_t, cuda::thread_scope_device> ref{*d_output};
     ref.fetch_add(total, cuda::std::memory_order_relaxed);
   }
 }
diff --git a/cpp/src/strings/split/split.cuh b/cpp/src/strings/split/split.cuh
index ae3c0b3aa12..23614ac0733 100644
--- a/cpp/src/strings/split/split.cuh
+++ b/cpp/src/strings/split/split.cuh
@@ -327,7 +327,7 @@ CUDF_KERNEL void count_delimiters_kernel(Tokenizer tokenizer,
   auto const total = block_reduce(temp_storage).Reduce(count, cub::Sum());
 
   if ((lane_idx == 0) && (total > 0)) {
-    cuda::atomic_ref<int64_t, cuda::thread_scope_block> ref{*d_output};
+    cuda::atomic_ref<int64_t, cuda::thread_scope_device> ref{*d_output};
     ref.fetch_add(total, cuda::std::memory_order_relaxed);
   }
 }
diff --git a/cpp/src/text/tokenize.cu b/cpp/src/text/tokenize.cu
index 25406bce759..3ce6064d9c2 100644
--- a/cpp/src/text/tokenize.cu
+++ b/cpp/src/text/tokenize.cu
@@ -121,7 +121,7 @@ CUDF_KERNEL void count_characters(uint8_t const* d_chars, int64_t chars_bytes, i
   auto const total = block_reduce(temp_storage).Reduce(count, cub::Sum());
 
   if ((lane_idx == 0) && (total > 0)) {
-    cuda::atomic_ref<int64_t, cuda::thread_scope_block> ref{*d_output};
+    cuda::atomic_ref<int64_t, cuda::thread_scope_device> ref{*d_output};
     ref.fetch_add(total, cuda::std::memory_order_relaxed);
   }
 }