diff --git a/test/distributed/test_c10d_nccl.py b/test/distributed/test_c10d_nccl.py
index f1691eeb877b..5849190e619c 100644
--- a/test/distributed/test_c10d_nccl.py
+++ b/test/distributed/test_c10d_nccl.py
@@ -285,10 +285,9 @@ class ProcessGroupNCCLGroupTest(MultiProcessTestCase):
     def setUp(self):
         super().setUp()
 
-        # These tests are expected to throw SIGABRT(6); adding the negative sign
-        # bc the test return code is actually -6
+        # These tests are expected to throw SIGABRT(6);
         # But if we are in Sandcastle, `skip_but_pass_in_sandcastle` would return 0.
-        TEST_NAN_ASSERT_RETURN = 0 if IS_SANDCASTLE else -signal.SIGABRT
+        TEST_NAN_ASSERT_RETURN = 0 if IS_SANDCASTLE else signal.SIGABRT
         self.special_return_code_checks = {
             self.test_nan_assert_float16.__wrapped__: TEST_NAN_ASSERT_RETURN,
             self.test_nan_assert_float32.__wrapped__: TEST_NAN_ASSERT_RETURN,
@@ -485,7 +484,7 @@ class ProcessGroupNCCLGroupTest(MultiProcessTestCase):
     @requires_nccl()
     @skip_but_pass_in_sandcastle_if(
         # skip for cu126 as well due to https://github.com/pytorch/pytorch/issues/153479
-        not (TEST_MULTIGPU and CUDA_12_AND_ABOVE and False),
+        not (TEST_MULTIGPU and CUDA_12_AND_ABOVE),
         "NCCL test requires 2+ GPUs and Device side assert could cause unexpected errors in lower versions of CUDA",
     )
     @parametrize(
@@ -539,10 +538,15 @@ class ProcessGroupNCCLGroupTest(MultiProcessTestCase):
         backend._set_enable_nan_check(False)
         # Note: using all-gather here bc some NCCL/SM version does not support
         # FP8 reduction
-        pg._allgather_base(output, nan_tensor)
+        # temporarily skip due to https://github.com/pytorch/pytorch/issues/153479
+        # pg._allgather_base(output, nan_tensor)
 
         backend._set_enable_nan_check(True)
-        pg._allgather_base(output, nan_tensor)
+        try:
+            pg._allgather_base(output, nan_tensor)
+        except Exception:
+            sys.exit(signal.SIGABRT)
+
         dist.destroy_process_group()
 
         # reset env