diff --git a/tests/unit/test_fmha_v3_stage_c.py b/tests/unit/test_fmha_v3_stage_c.py
index c868ec98..2aad1970 100644
--- a/tests/unit/test_fmha_v3_stage_c.py
+++ b/tests/unit/test_fmha_v3_stage_c.py
@@ -190,11 +190,11 @@ class FmhaV3StageCMulti:
         )
         epilog_sync_bar.arrive_and_wait()
 
+        # All warps in the epilogue group do the TMA store together
         c_buffer = 0
-        if warp_idx == self.epilogue_warp_id[0]:
-            cute.copy(tma_c, tCsC[(None, c_buffer)], tCgC_tma[(None, 0, 0, 0, 0, 0, 0)])
-            cute.arch.cp_async_bulk_commit_group()
-            cute.arch.cp_async_bulk_wait_group(0, read=True)
+        cute.copy(tma_c, tCsC[(None, c_buffer)], tCgC_tma[(None, 0, 0, 0, 0, 0, 0)])
+        cute.arch.cp_async_bulk_commit_group()
+        cute.arch.cp_async_bulk_wait_group(0, read=True)
         epilog_sync_bar.arrive_and_wait()
 
     @cute.kernel