diff --git a/tests/unit/test_fmha_ts_hd16.cu b/tests/unit/test_fmha_ts_hd16.cu
index 5dc8331c..070bd33c 100644
--- a/tests/unit/test_fmha_ts_hd16.cu
+++ b/tests/unit/test_fmha_ts_hd16.cu
@@ -119,11 +119,11 @@ test_fmha_ts(const bf16_t* q, const bf16_t* k, const bf16_t* v,
     __syncthreads();
     */
 
-    // Zero O TMEM region
+    // Zero O TMEM region at tb (overwriting P — same as isolated test)
     if (wid == 0) {
-        for (int n = 0; n < TMEM_O / 8; n++) {
+        for (int n = 0; n < 2; n++) {  // 16 cols / 8 = 2 iterations
             float z0=0,z1=0,z2=0,z3=0,z4=0,z5=0,z6=0,z7=0;
-            asm volatile("tcgen05.st.sync.aligned.32x32b.x8.b32 [%0],{%1,%2,%3,%4,%5,%6,%7,%8};" :: "r"(tb_o+n*8),"f"(z0),"f"(z1),"f"(z2),"f"(z3),"f"(z4),"f"(z5),"f"(z6),"f"(z7));
+            asm volatile("tcgen05.st.sync.aligned.32x32b.x8.b32 [%0],{%1,%2,%3,%4,%5,%6,%7,%8};" :: "r"(tb+n*8),"f"(z0),"f"(z1),"f"(z2),"f"(z3),"f"(z4),"f"(z5),"f"(z6),"f"(z7));
         }
         tmem_fence_store();
     }
@@ -139,7 +139,7 @@ test_fmha_ts(const bf16_t* q, const bf16_t* k, const bf16_t* v,
         uint64_t dv = make_umma_desc_kmajor_none(__cvta_generic_to_shared(sV_base), 16);
         uint32_t tmem_a = tb;  // P's first 16 columns
 
-        if (tid == 0) umma_ts_f16(tb_o, tmem_a, dv, idesc_pv, false);  // no accumulate for first tile
+        if (tid == 0) umma_ts_f16(tb, tmem_a, dv, idesc_pv, false);  // Write C to tb (overwriting P)
         asm volatile("tcgen05.fence::after_thread_sync;" ::: "memory");
         __syncthreads();
     }
@@ -151,7 +151,7 @@ test_fmha_ts(const bf16_t* q, const bf16_t* k, const bf16_t* v,
         float o_vals[HD];
         for (int n = 0; n < HD / 8; n++) {
             float tmp[8];
-            asm volatile("tcgen05.ld.sync.aligned.32x32b.x8.b32 {%0,%1,%2,%3,%4,%5,%6,%7},[%8];" : "=f"(tmp[0]),"=f"(tmp[1]),"=f"(tmp[2]),"=f"(tmp[3]),"=f"(tmp[4]),"=f"(tmp[5]),"=f"(tmp[6]),"=f"(tmp[7]) : "r"(tb_o + n*8));
+            asm volatile("tcgen05.ld.sync.aligned.32x32b.x8.b32 {%0,%1,%2,%3,%4,%5,%6,%7},[%8];" : "=f"(tmp[0]),"=f"(tmp[1]),"=f"(tmp[2]),"=f"(tmp[3]),"=f"(tmp[4]),"=f"(tmp[5]),"=f"(tmp[6]),"=f"(tmp[7]) : "r"(tb + n*8));
             asm volatile("tcgen05.wait::ld.sync.aligned;");
             if (lane == 0) for (int c=0;c<8;c++) o_vals[n*8+c] = tmp[c] * 2.0f;  // Undo MMA 0.5 scale
         }