diff --git a/numba/cuda/cudadrv/devicearray.py b/numba/cuda/cudadrv/devicearray.py
index 651f06bf698..21adaa94cf3 100644
--- a/numba/cuda/cudadrv/devicearray.py
+++ b/numba/cuda/cudadrv/devicearray.py
@@ -895,6 +895,8 @@ def check_array_compatibility(ary1, ary2):
     if ary1sq.shape != ary2sq.shape:
         raise ValueError('incompatible shape: %s vs. %s' %
                          (ary1.shape, ary2.shape))
-    if ary1sq.strides != ary2sq.strides:
+    # We check strides only if the size is nonzero, because strides are
+    # irrelevant (and can differ) for zero-length copies.
+    if ary1.size and ary1sq.strides != ary2sq.strides:
         raise ValueError('incompatible strides: %s vs. %s' %
                          (ary1.strides, ary2.strides))
diff --git a/numba/cuda/tests/cudadrv/test_cuda_ndarray.py b/numba/cuda/tests/cudadrv/test_cuda_ndarray.py
index d84d297ebca..1e674df7aca 100644
--- a/numba/cuda/tests/cudadrv/test_cuda_ndarray.py
+++ b/numba/cuda/tests/cudadrv/test_cuda_ndarray.py
@@ -420,6 +420,41 @@ def test_bug6697(self):
         got = np.asarray(dary)
         self.assertEqual(got.dtype, dary.dtype)
 
+    @skip_on_cudasim('DeviceNDArray class not present in simulator')
+    def test_issue_8477(self):
+        # Ensure that we can copy a zero-length device array to a zero-length
+        # host array when the strides of the device and host arrays differ -
+        # this should be possible because the strides are irrelevant when the
+        # length is zero. For more info see
+        # https://github.com/numba/numba/issues/8477.
+
+        # Create a device array with shape (0,) and strides (8,)
+        dev_array = devicearray.DeviceNDArray(shape=(0,), strides=(8,),
+                                              dtype=np.int8)
+
+        # Create a host array with shape (0,) and strides (0,)
+        host_array = np.ndarray(shape=(0,), strides=(0,), dtype=np.int8)
+
+        # Sanity check for this test - ensure our destination has the strides
+        # we expect, because strides can be ignored in some cases by the
+        # ndarray constructor - checking here ensures that we haven't failed to
+        # account for unexpected behaviour across different versions of NumPy
+        self.assertEqual(host_array.strides, (0,))
+
+        # Ensure that the copy succeeds in both directions
+        dev_array.copy_to_host(host_array)
+        dev_array.copy_to_device(host_array)
+
+        # Ensure that a device-to-device copy also succeeds when the strides
+        # differ - one way of doing this is to copy the host array across and
+        # use that for copies in both directions.
+        dev_array_from_host = cuda.to_device(host_array)
+        self.assertEqual(dev_array_from_host.shape, (0,))
+        self.assertEqual(dev_array_from_host.strides, (0,))
+
+        dev_array.copy_to_device(dev_array_from_host)
+        dev_array_from_host.copy_to_device(dev_array)
+
 
 class TestRecarray(CUDATestCase):
     def test_recarray(self):