diff --git a/tests/array_interoperability_test.py b/tests/array_interoperability_test.py
index 0eca900dc54e..ba07ff344c85 100644
--- a/tests/array_interoperability_test.py
+++ b/tests/array_interoperability_test.py
@@ -74,7 +74,7 @@ def setUp(self):
   def testJaxRoundTrip(self, shape, dtype, take_ownership, gpu):
     rng = jtu.rand_default(self.rng())
     np = rng(shape, dtype)
-    if gpu and jax.default_backend() == "cpu":
+    if gpu and jax.test_device_matches(["cpu"]):
       raise unittest.SkipTest("Skipping GPU test case on CPU")
     device = jax.devices("gpu" if gpu else "cpu")[0]
     x = jax.device_put(np, device)
@@ -180,7 +180,7 @@ def testNumpyToJax(self, shape, dtype):
     dtype=numpy_dtypes,
   )
   @unittest.skipIf(numpy_version < (1, 23, 0), "Requires numpy 1.23 or newer")
-  @jtu.skip_on_devices("gpu") #NumPy only accepts cpu DLPacks
+  @jtu.run_on_devices("cpu") # NumPy only accepts cpu DLPacks
   def testJaxToNumpy(self, shape, dtype):
     rng = jtu.rand_default(self.rng())
     x_jax = jnp.array(rng(shape, dtype))
@@ -192,7 +192,7 @@ class CudaArrayInterfaceTest(jtu.JaxTestCase):
 
   def setUp(self):
     super().setUp()
-    if not jtu.test_device_matches(["gpu"]):
+    if not jtu.test_device_matches(["cuda"]):
       self.skipTest("__cuda_array_interface__ is only supported on GPU")
 
   @jtu.sample_product(
diff --git a/tests/debugging_primitives_test.py b/tests/debugging_primitives_test.py
index 89fe88899575..42a0872737eb 100644
--- a/tests/debugging_primitives_test.py
+++ b/tests/debugging_primitives_test.py
@@ -98,10 +98,8 @@ def f(x):
       jax.effects_barrier()
     self.assertEqual(output(), "x: 2\n")
 
+  @jtu.device_supports_buffer_donation()
   def test_can_stage_out_debug_print_with_donate_argnums(self):
-    if jax.default_backend() not in {"gpu", "tpu"}:
-      raise unittest.SkipTest("Donate argnums not supported.")
-
     def f(x, y):
       debug_print('x: {x}', x=x)
       return x + y
@@ -120,10 +118,8 @@ def f(x):
       jax.effects_barrier()
     self.assertEqual(output(), "x: 2\n")
 
+  @jtu.device_supports_buffer_donation()
   def test_can_stage_out_ordered_print_with_donate_argnums(self):
-    if jax.default_backend() not in {"gpu", "tpu"}:
-      raise unittest.SkipTest("Donate argnums not supported.")
-
     def f(x, y):
       debug_print('x: {x}', x=x, ordered=True)
       return x + y
@@ -133,10 +129,8 @@ def f(x, y):
       jax.effects_barrier()
     self.assertEqual(output(), "x: 2\n")
 
+  @jtu.device_supports_buffer_donation()
   def test_can_stage_out_prints_with_donate_argnums(self):
-    if jax.default_backend() not in {"gpu", "tpu"}:
-      raise unittest.SkipTest("Donate argnums not supported.")
-
     def f(x, y):
       debug_print('x: {x}', x=x, ordered=True)
       debug_print('x: {x}', x=x)
diff --git a/tests/lax_numpy_test.py b/tests/lax_numpy_test.py
index 51b1710ea27f..7940cf317969 100644
--- a/tests/lax_numpy_test.py
+++ b/tests/lax_numpy_test.py
@@ -1275,7 +1275,7 @@ def testTrimZeros(self, a_shape, dtype, trim):
   def testPoly(self, a_shape, dtype, rank):
     if dtype in (np.float16, jnp.bfloat16, np.int16):
       self.skipTest(f"{dtype} gets promoted to {np.float16}, which is not supported.")
-    elif rank == 2 and jtu.test_device_matches(["tpu", "gpu"]):
+    elif rank == 2 and not jtu.test_device_matches(["cpu"]):
       self.skipTest("Nonsymmetric eigendecomposition is only implemented on the CPU backend.")
     rng = jtu.rand_default(self.rng())
     tol = { np.int8: 2e-3, np.int32: 1e-3, np.float32: 1e-3, np.float64: 1e-6 }
@@ -1914,7 +1914,7 @@ def np_fun(x, y):
     xshape=one_dim_array_shapes,
     yshape=one_dim_array_shapes,
   )
-  @jtu.skip_on_devices("gpu", "tpu", "rocm")  # backends don't support all dtypes.
+  @jtu.skip_on_devices("cuda", "tpu", "rocm")  # backends don't support all dtypes.
   def testConvolutionsPreferredElementType(self, xshape, yshape, dtype, mode, op):
     jnp_op = getattr(jnp, op)
     np_op = getattr(np, op)
diff --git a/tests/linalg_test.py b/tests/linalg_test.py
index 13abf2811049..9404b9fee11b 100644
--- a/tests/linalg_test.py
+++ b/tests/linalg_test.py
@@ -210,7 +210,7 @@ def testIssue1213(self):
   )
   # TODO(phawkins): enable when there is an eigendecomposition implementation
   # for GPU/TPU.
-  @jtu.skip_on_devices("gpu", "tpu")
+  @jtu.run_on_devices("cpu")
   def testEig(self, shape, dtype, compute_left_eigenvectors,
               compute_right_eigenvectors):
     rng = jtu.rand_default(self.rng())
@@ -252,7 +252,7 @@ def check_left_eigenvectors(a, w, vl):
   )
   # TODO(phawkins): enable when there is an eigendecomposition implementation
   # for GPU/TPU.
-  @jtu.skip_on_devices("gpu", "tpu")
+  @jtu.run_on_devices("cpu")
   def testEigvalsGrad(self, shape, dtype):
     # This test sometimes fails for large matrices. I (@j-towns) suspect, but
     # haven't checked, that might be because of perturbations causing the
@@ -271,7 +271,7 @@ def testEigvalsGrad(self, shape, dtype):
   )
   # TODO: enable when there is an eigendecomposition implementation
   # for GPU/TPU.
-  @jtu.skip_on_devices("gpu", "tpu")
+  @jtu.run_on_devices("cpu")
   def testEigvals(self, shape, dtype):
     rng = jtu.rand_default(self.rng())
     args_maker = lambda: [rng(shape, dtype)]
@@ -280,7 +280,7 @@ def testEigvals(self, shape, dtype):
     w2 = jnp.linalg.eigvals(a)
     self.assertAllClose(w1, w2, rtol={np.complex64: 1e-5, np.complex128: 1e-14})
 
-  @jtu.skip_on_devices("gpu", "tpu")
+  @jtu.run_on_devices("cpu")
   def testEigvalsInf(self):
     # https://github.com/google/jax/issues/2661
     x = jnp.array([[jnp.inf]])
@@ -290,7 +290,7 @@ def testEigvalsInf(self):
     shape=[(1, 1), (4, 4), (5, 5)],
     dtype=float_types + complex_types,
   )
-  @jtu.skip_on_devices("gpu", "tpu")
+  @jtu.run_on_devices("cpu")
   def testEigBatching(self, shape, dtype):
     rng = jtu.rand_default(self.rng())
     shape = (10,) + shape
@@ -688,7 +688,7 @@ def testNumpyQrModes(self, shape, dtype, mode):
   )
   @jax.default_matmul_precision("float32")
   def testQr(self, shape, dtype, full_matrices):
-    if (jtu.test_device_matches(["gpu"]) and
+    if (jtu.test_device_matches(["cuda"]) and
         _is_required_cuda_version_satisfied(12000)):
       self.skipTest("Triggers a bug in cuda-12 b/287345077")
     rng = jtu.rand_default(self.rng())
@@ -1287,7 +1287,7 @@ def testTriangularSolveGradPrecision(self):
     dtype=int_types + float_types + complex_types
   )
   def testExpm(self, n, batch_size, dtype):
-    if (jtu.test_device_matches(["gpu"]) and
+    if (jtu.test_device_matches(["cuda"]) and
         _is_required_cuda_version_satisfied(12000)):
       self.skipTest("Triggers a bug in cuda-12 b/287345077")
 
@@ -1357,7 +1357,7 @@ def reference_fn(a, taus):
       dtype=float_types + complex_types,
       calc_q=[False, True],
   )
-  @jtu.skip_on_devices("gpu", "tpu")
+  @jtu.run_on_devices("cpu")
   def testHessenberg(self, shape, dtype, calc_q):
     rng = jtu.rand_default(self.rng())
     jsp_func = partial(jax.scipy.linalg.hessenberg, calc_q=calc_q)
@@ -1514,7 +1514,7 @@ def expm(x):
     shape=[(4, 4), (15, 15), (50, 50), (100, 100)],
     dtype=float_types + complex_types,
   )
-  @jtu.skip_on_devices("gpu", "tpu")
+  @jtu.run_on_devices("cpu")
   def testSchur(self, shape, dtype):
       rng = jtu.rand_default(self.rng())
       args_maker = lambda: [rng(shape, dtype)]
@@ -1526,7 +1526,7 @@ def testSchur(self, shape, dtype):
     shape=[(1, 1), (4, 4), (15, 15), (50, 50), (100, 100)],
     dtype=float_types + complex_types,
   )
-  @jtu.skip_on_devices("gpu", "tpu")
+  @jtu.run_on_devices("cpu")
   def testRsf2csf(self, shape, dtype):
       rng = jtu.rand_default(self.rng())
       args_maker = lambda: [rng(shape, dtype), rng(shape, dtype)]
@@ -1542,7 +1542,7 @@ def testRsf2csf(self, shape, dtype):
   )
   # funm uses jax.scipy.linalg.schur which is implemented for a CPU
   # backend only, so tests on GPU and TPU backends are skipped here
-  @jtu.skip_on_devices("gpu", "tpu")
+  @jtu.run_on_devices("cpu")
   def testFunm(self, shape, dtype, disp):
       def func(x):
         return x**-2.718
@@ -1558,7 +1558,7 @@ def func(x):
     shape=[(4, 4), (15, 15), (50, 50), (100, 100)],
     dtype=float_types + complex_types,
   )
-  @jtu.skip_on_devices("gpu", "tpu")
+  @jtu.run_on_devices("cpu")
   def testSqrtmPSDMatrix(self, shape, dtype):
     # Checks against scipy.linalg.sqrtm when the principal square root
     # is guaranteed to be unique (i.e no negative real eigenvalue)
@@ -1581,7 +1581,7 @@ def testSqrtmPSDMatrix(self, shape, dtype):
     shape=[(4, 4), (15, 15), (50, 50), (100, 100)],
     dtype=float_types + complex_types,
   )
-  @jtu.skip_on_devices("gpu", "tpu")
+  @jtu.run_on_devices("cpu")
   def testSqrtmGenMatrix(self, shape, dtype):
     rng = jtu.rand_default(self.rng())
     arg = rng(shape, dtype)
@@ -1600,7 +1600,7 @@ def testSqrtmGenMatrix(self, shape, dtype):
     ],
     dtype=float_types + complex_types,
   )
-  @jtu.skip_on_devices("gpu", "tpu")
+  @jtu.run_on_devices("cpu")
   def testSqrtmEdgeCase(self, diag, expected, dtype):
     """
     Tests the zero numerator condition
@@ -1773,7 +1773,7 @@ def test_tridiagonal_solve(self, dtype):
     shape=[(4, 4), (15, 15), (50, 50), (100, 100)],
     dtype=float_types + complex_types,
   )
-  @jtu.skip_on_devices("gpu", "tpu")
+  @jtu.run_on_devices("cpu")
   def testSchur(self, shape, dtype):
     rng = jtu.rand_default(self.rng())
     args_maker = lambda: [rng(shape, dtype)]
@@ -1785,7 +1785,7 @@ def testSchur(self, shape, dtype):
     shape=[(2, 2), (4, 4), (15, 15), (50, 50), (100, 100)],
     dtype=float_types + complex_types,
   )
-  @jtu.skip_on_devices("gpu", "tpu")
+  @jtu.run_on_devices("cpu")
   def testSchurBatching(self, shape, dtype):
     rng = jtu.rand_default(self.rng())
     batch_size = 10
diff --git a/tests/polynomial_test.py b/tests/polynomial_test.py
index cdbf19743803..ccba4c2ef11f 100644
--- a/tests/polynomial_test.py
+++ b/tests/polynomial_test.py
@@ -72,7 +72,7 @@ def assertSetsAllClose(self, x, y, rtol=None, atol=None, check_dtypes=True):
     trailing=[0, 2],
   )
   # TODO(phawkins): no nonsymmetric eigendecomposition implementation on GPU.
-  @jtu.skip_on_devices("gpu", "tpu")
+  @jtu.run_on_devices("cpu")
   def testRoots(self, dtype, length, leading, trailing):
     rng = jtu.rand_some_zero(self.rng())
 
@@ -98,7 +98,7 @@ def np_fun(arg):
     trailing=[0, 2],
   )
   # TODO(phawkins): no nonsymmetric eigendecomposition implementation on GPU.
-  @jtu.skip_on_devices("gpu", "tpu")
+  @jtu.run_on_devices("cpu")
   def testRootsNoStrip(self, dtype, length, leading, trailing):
     rng = jtu.rand_some_zero(self.rng())
 
diff --git a/tests/sparse_test.py b/tests/sparse_test.py
index 71f765117c90..17c55a7df2a3 100644
--- a/tests/sparse_test.py
+++ b/tests/sparse_test.py
@@ -1958,7 +1958,7 @@ def test_bcsr_matmul(self, lhs_shape, lhs_dtype, rhs_shape, rhs_dtype):
   @jax.default_matmul_precision("float32")
   @jtu.ignore_warning(category=sparse.CuSparseEfficiencyWarning)
   def test_bcoo_matmul(self, lhs_shape, lhs_dtype, rhs_shape, rhs_dtype):
-    if (jtu.test_device_matches(["gpu"]) and
+    if (jtu.test_device_matches(["cuda"]) and
         _is_required_cuda_version_satisfied(12000)):
       raise unittest.SkipTest("Triggers a bug in cuda-12 b/287344632")
 
@@ -2777,8 +2777,7 @@ class SparseSolverTest(sptu.SparseTestCase):
     reorder=[0, 1, 2, 3],
     dtype=jtu.dtypes.floating + jtu.dtypes.complex,
   )
-  @jtu.run_on_devices("cpu", "gpu")
-  @jtu.skip_on_devices("rocm")  # test n gpu requires cusolver
+  @jtu.run_on_devices("cpu", "cuda")
   def test_sparse_qr_linear_solver(self, size, reorder, dtype):
     if jtu.test_device_matches(["cuda"]) and not GPU_LOWERING_ENABLED:
       raise unittest.SkipTest('test requires cusparse/cusolver')
@@ -2805,8 +2804,7 @@ def sparse_solve(data, indices, indptr, b):
     size=[10, 20, 50],
     dtype=jtu.dtypes.floating,
   )
-  @jtu.run_on_devices("cpu", "gpu")
-  @jtu.skip_on_devices("rocm")  # test requires cusolver
+  @jtu.run_on_devices("cpu", "cuda")
   def test_sparse_qr_linear_solver_grads(self, size, dtype):
     if jtu.test_device_matches(["cuda"]) and not GPU_LOWERING_ENABLED:
       raise unittest.SkipTest('test requires cusparse/cusolver')
diff --git a/tests/xmap_test.py b/tests/xmap_test.py
index dc424c7fdbb4..76fa9c22e3d4 100644
--- a/tests/xmap_test.py
+++ b/tests/xmap_test.py
@@ -459,7 +459,7 @@ def f(x, y):
       self.assertAllClose(f_mapped(x, x), expected)
 
   @jtu.with_and_without_mesh
-  @jtu.run_on_devices("gpu", "tpu")  # In/out aliasing not supported on CPU.
+  @jtu.device_supports_buffer_donation()  # In/out aliasing not supported on CPU
   def testBufferDonation(self, mesh, axis_resources):
     shard = lambda x: x
     if axis_resources:
@@ -476,7 +476,7 @@ def testBufferDonation(self, mesh, axis_resources):
     self.assertNotDeleted(y)
     self.assertDeleted(x)
 
-  @jtu.run_on_devices("gpu", "tpu")  # In/out aliasing not supported on CPU.
+  @jtu.device_supports_buffer_donation()  # In/out aliasing not supported on CPU
   @jtu.with_mesh([('x', 2)])
   @jtu.ignore_warning(category=UserWarning,  # SPMD test generates warning.
                       message="Some donated buffers were not usable*")