diff --git a/cpp/src/neighbors/mg/snmg.cuh b/cpp/src/neighbors/mg/snmg.cuh
index 057a3e8272..a6b057693b 100644
--- a/cpp/src/neighbors/mg/snmg.cuh
+++ b/cpp/src/neighbors/mg/snmg.cuh
@@ -355,11 +355,22 @@ void sharded_search_with_direct_merge(
                d_trans.view(),
                raft::make_host_vector_view<const searchIdxT>(h_trans.data(), index.num_ranks_));
 
+    // Results from each rank are packed using the current batch size. Use matching logical views
+    // so the final partial batch is merged with the same per-rank stride used above.
+    auto in_distances_batch = raft::make_device_matrix_view<const float, int64_t, row_major>(
+      in_distances.data_handle(), index.num_ranks_ * n_rows_of_current_batch, n_neighbors);
+    auto in_neighbors_batch = raft::make_device_matrix_view<const searchIdxT, int64_t, row_major>(
+      in_neighbors.data_handle(), index.num_ranks_ * n_rows_of_current_batch, n_neighbors);
+    auto out_distances_batch = raft::make_device_matrix_view<float, int64_t, row_major>(
+      out_distances.data_handle(), n_rows_of_current_batch, n_neighbors);
+    auto out_neighbors_batch = raft::make_device_matrix_view<searchIdxT, int64_t, row_major>(
+      out_neighbors.data_handle(), n_rows_of_current_batch, n_neighbors);
+
     knn_merge_parts(root_handle_,
-                    in_distances.view(),
-                    in_neighbors.view(),
-                    out_distances.view(),
-                    out_neighbors.view(),
+                    in_distances_batch,
+                    in_neighbors_batch,
+                    out_distances_batch,
+                    out_neighbors_batch,
                     d_trans.view());
 
     raft::copy(