diff --git a/tests/unit/test_umma_qk.cu b/tests/unit/test_umma_qk.cu
index 648618dd..d92a6563 100644
--- a/tests/unit/test_umma_qk.cu
+++ b/tests/unit/test_umma_qk.cu
@@ -78,12 +78,12 @@ test_umma_qk_hd16(
     }
     __syncthreads();
 
-    // Zero TMEM — test with 32 columns first
+    // Zero TMEM — test without fence
     if (wid == 0) {
-        for (int col = 0; col < 2; col++) {
+        for (int col = 0; col < 128; col++) {
             tmem_store(tmem_base + col, 0, 0, 0, 0);
         }
-        tmem_fence_store();
+        // tmem_fence_store();  // skip fence
     }
     __syncthreads();