diff --git a/CHANGELOG.md b/CHANGELOG.md
index 4092e04d5765b..46bb8106bf80b 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -600,6 +600,9 @@ The format is based on [Keep a Changelog](http://keepachangelog.com/en/1.0.0/).
 - Deprecated passing only the callback state to `Callback.on_load_checkpoint(callback_state)` in favor of passing the callback state to `Callback.load_state_dict` and in 1.8, passing the entire checkpoint dictionary to `Callback.on_load_checkpoint(checkpoint)` ([#11887](https://github.com/PyTorchLightning/pytorch-lightning/pull/11887))
 
 
+- Deprecated `Trainer.gpus` in favor of `Trainer.device_ids` or `Trainer.num_devices` ([#12436](https://github.com/PyTorchLightning/pytorch-lightning/pull/12436))
+
+
 ### Removed
 
 - Removed deprecated parameter `method` in `pytorch_lightning.utilities.model_helpers.is_overridden` ([#10507](https://github.com/PyTorchLightning/pytorch-lightning/pull/10507))
diff --git a/pytorch_lightning/callbacks/gpu_stats_monitor.py b/pytorch_lightning/callbacks/gpu_stats_monitor.py
index 607dc4cf0efbd..36b4006e37a2a 100644
--- a/pytorch_lightning/callbacks/gpu_stats_monitor.py
+++ b/pytorch_lightning/callbacks/gpu_stats_monitor.py
@@ -128,8 +128,8 @@ def setup(self, trainer: "pl.Trainer", pl_module: "pl.LightningModule", stage: O
 
         if trainer.strategy.root_device.type != "cuda":
             raise MisconfigurationException(
-                "You are using GPUStatsMonitor but are not running on GPU"
-                f" since gpus attribute in Trainer is set to {trainer.gpus}."
+                "You are using GPUStatsMonitor but are not running on GPU."
+                f" The root device type is {trainer.strategy.root_device.type}."
             )
 
         # The logical device IDs for selected devices
diff --git a/pytorch_lightning/trainer/trainer.py b/pytorch_lightning/trainer/trainer.py
index bcda0764ff11f..e5168dfef4c83 100644
--- a/pytorch_lightning/trainer/trainer.py
+++ b/pytorch_lightning/trainer/trainer.py
@@ -2195,6 +2195,10 @@ def scaler(self) -> Optional[Any]:
 
     @property
     def gpus(self) -> Optional[Union[List[int], str, int]]:
+        rank_zero_deprecation(
+            "`Trainer.gpus` was deprecated in v1.6 and will be removed in v1.8."
+            " Please use `Trainer.num_devices` or `Trainer.device_ids` to get device information instead."
+        )
         return self._accelerator_connector.gpus
 
     @property
diff --git a/tests/accelerators/test_accelerator_connector.py b/tests/accelerators/test_accelerator_connector.py
index 6edbb0eefff90..5a99c452b299a 100644
--- a/tests/accelerators/test_accelerator_connector.py
+++ b/tests/accelerators/test_accelerator_connector.py
@@ -554,7 +554,7 @@ def test_accelerator_gpu_with_devices(devices, plugin):
 def test_accelerator_auto_with_devices_gpu():
     trainer = Trainer(accelerator="auto", devices=1)
     assert isinstance(trainer.accelerator, GPUAccelerator)
-    assert trainer.gpus == 1
+    assert trainer.num_devices == 1
 
 
 def test_validate_accelerator_and_devices():
@@ -946,8 +946,8 @@ def test_devices_auto_choice_cpu(
 @mock.patch("torch.cuda.device_count", return_value=2)
 def test_devices_auto_choice_gpu(is_gpu_available_mock, device_count_mock):
     trainer = Trainer(accelerator="auto", devices="auto")
+    assert isinstance(trainer.accelerator, GPUAccelerator)
     assert trainer.num_devices == 2
-    assert trainer.gpus == 2
 
 
 @pytest.mark.parametrize(
diff --git a/tests/deprecated_api/test_remove_1-8.py b/tests/deprecated_api/test_remove_1-8.py
index 2157dd2ad38fa..a7cabf1794a8f 100644
--- a/tests/deprecated_api/test_remove_1-8.py
+++ b/tests/deprecated_api/test_remove_1-8.py
@@ -1106,3 +1106,23 @@ def on_save_checkpoint(self, trainer, pl_module, checkpoint):
 
     trainer.callbacks = [TestCallbackSaveHookOverride()]
     trainer.save_checkpoint(tmpdir + "/pathok.ckpt")
+
+
+@pytest.mark.parametrize(
+    "trainer_kwargs",
+    [
+        {"accelerator": "gpu", "devices": 2},
+        {"accelerator": "gpu", "devices": [0, 2]},
+        {"accelerator": "gpu", "devices": "2"},
+        {"accelerator": "gpu", "devices": "0,"},
+    ],
+)
+def test_trainer_gpus(monkeypatch, trainer_kwargs):
+    monkeypatch.setattr(torch.cuda, "is_available", lambda: True)
+    monkeypatch.setattr(torch.cuda, "device_count", lambda: 4)
+    trainer = Trainer(**trainer_kwargs)
+    with pytest.deprecated_call(
+        match="`Trainer.gpus` was deprecated in v1.6 and will be removed in v1.8."
+        " Please use `Trainer.num_devices` or `Trainer.device_ids` to get device information instead."
+    ):
+        assert trainer.gpus == trainer_kwargs["devices"]
diff --git a/tests/models/test_gpu.py b/tests/models/test_gpu.py
index e483f3e627b70..e805b1663f681 100644
--- a/tests/models/test_gpu.py
+++ b/tests/models/test_gpu.py
@@ -23,6 +23,7 @@
 import tests.helpers.pipelines as tpipes
 import tests.helpers.utils as tutils
 from pytorch_lightning import Trainer
+from pytorch_lightning.accelerators import CPUAccelerator, GPUAccelerator
 from pytorch_lightning.plugins.environments import TorchElasticEnvironment
 from pytorch_lightning.utilities import device_parser
 from pytorch_lightning.utilities.exceptions import MisconfigurationException
@@ -192,10 +193,16 @@ def test_torchelastic_gpu_parsing(mocked_device_count, mocked_is_available, gpus
     sanitizing the gpus as only one of the GPUs is visible."""
     trainer = Trainer(gpus=gpus)
     assert isinstance(trainer._accelerator_connector.cluster_environment, TorchElasticEnvironment)
-    assert trainer.gpus == gpus
     # when use gpu
     if device_parser.parse_gpu_ids(gpus) is not None:
+        assert isinstance(trainer.accelerator, GPUAccelerator)
+        assert trainer.num_devices == len(gpus) if isinstance(gpus, list) else gpus
         assert trainer.device_ids == device_parser.parse_gpu_ids(gpus)
+    # fall back to cpu
+    else:
+        assert isinstance(trainer.accelerator, CPUAccelerator)
+        assert trainer.num_devices == 1
+        assert trainer.device_ids == [0]
 
 
 @RunIf(min_gpus=1)