diff --git a/src/opr/impl/atlas_runtime_op.cpp b/src/opr/impl/atlas_runtime_op.cpp
index 7853545a..46d3e437 100644
--- a/src/opr/impl/atlas_runtime_op.cpp
+++ b/src/opr/impl/atlas_runtime_op.cpp
@@ -278,6 +278,9 @@ void AtlasRuntimeOpr::scn_do_execute() {
         for (size_t i = 0; i < output().size(); i++) {
             auto output_size = aclmdlGetOutputSizeByIndex(m_model_desc, i);
             auto ovar = output(i);
+            output_size = std::max<size_t>(
+                    output_size,
+                    ovar->dtype().size(ovar->shape().total_nr_elems()));
             ovar->shape_alloc(ovar->shape(), output_size);
         }
     }
diff --git a/src/opr/test/atlas_runtime_op.cpp b/src/opr/test/atlas_runtime_op.cpp
index 19ab0f24..b837da44 100644
--- a/src/opr/test/atlas_runtime_op.cpp
+++ b/src/opr/test/atlas_runtime_op.cpp
@@ -65,7 +65,7 @@ TEST(TestOprAtlas, Basic) {
 }
 
 TEST(TestOprAtlas, DynamicBatch) {
-    for (size_t batch : {1, 6}) {
+    for (size_t batch : {1, 6, 20}) {
         HostTensorGenerator<> gen;
         const auto& graph = ComputingGraph::make();
         const auto& host_x = gen({batch, 3, 16, 16});