diff --git a/pytorch_lightning/utilities/imports.py b/pytorch_lightning/utilities/imports.py
index caf978456c03d..eb64e03a559a9 100644
--- a/pytorch_lightning/utilities/imports.py
+++ b/pytorch_lightning/utilities/imports.py
@@ -87,7 +87,7 @@ def _compare_version(package: str, op, version) -> bool:
 _POPTORCH_AVAILABLE = _module_available("poptorch")
 _RICH_AVAILABLE = _module_available("rich")
 _TORCH_CPU_AMP_AVAILABLE = _compare_version(
-    "torch", operator.ge, "1.10.0dev20210501"
+    "torch", operator.ge, "1.10.dev20210902"
 )  # todo: swap to 1.10.0 once released
 _TORCH_BFLOAT_AVAILABLE = _compare_version(
     "torch", operator.ge, "1.10.0.dev20210902"
diff --git a/tests/models/test_amp.py b/tests/models/test_amp.py
index 8632b3b9eb6f6..c56cfb11f0164 100644
--- a/tests/models/test_amp.py
+++ b/tests/models/test_amp.py
@@ -184,10 +184,13 @@ def test_amp_gpu_ddp_slurm_managed(tmpdir):
 
 
 @pytest.mark.skipif(torch.cuda.is_available(), reason="test is restricted only on CPU")
-def test_cpu_model_with_amp(tmpdir):
-    """Make sure model trains on CPU."""
-    with pytest.raises(MisconfigurationException, match="AMP is only available on GPU"):
-        Trainer(precision=16)
+@RunIf(max_torch="1.9")
+@pytest.mark.parametrize("precision", [16, "bf16"])
+def test_cpu_model_with_amp(tmpdir, precision):
+    """Make sure exception is thrown on CPU when precision 16 is enabled on PyTorch 1.9 and lower."""
+
+    with pytest.raises(MisconfigurationException, match="AMP is only available on GPU for PyTorch 1.9"):
+        Trainer(precision=precision)
 
 
 @mock.patch("pytorch_lightning.plugins.precision.apex_amp.ApexMixedPrecisionPlugin.backward")