diff --git a/include/ck_tile/ops/gemm/pipeline/gemm_pipeline_ag_bg_cr_comp_v4.hpp b/include/ck_tile/ops/gemm/pipeline/gemm_pipeline_ag_bg_cr_comp_v4.hpp
index f22d2f599a..d448cdbb93 100644
--- a/include/ck_tile/ops/gemm/pipeline/gemm_pipeline_ag_bg_cr_comp_v4.hpp
+++ b/include/ck_tile/ops/gemm/pipeline/gemm_pipeline_ag_bg_cr_comp_v4.hpp
@@ -23,7 +23,8 @@ struct BaseGemmPipelineAgBgCrCompV4
 
     CK_TILE_HOST_DEVICE static constexpr bool BlockHasHotloop(index_t num_loop)
     {
-        return num_loop > PrefetchStages;
+        constexpr index_t HotLoopGlobalReads = 2;
+        return num_loop >= (HotLoopGlobalReads + PrefetchStages);
     }
 
     CK_TILE_HOST_DEVICE static constexpr TailNumber GetBlockLoopTailNum(index_t num_loop)