diff --git a/.github/workflows/test.yml b/.github/workflows/test.yml
index 3ace5c52f9..9d97541df2 100644
--- a/.github/workflows/test.yml
+++ b/.github/workflows/test.yml
@@ -584,7 +584,7 @@ jobs:
         key: metal
         deps: testing
     - name: some unit tests
-      run: METAL=1 RANGEIFY=1 python -m pytest -n=auto test/unit/test_winograd.py --durations=20
+      run: METAL=1 RANGEIFY=1 python -m pytest -n=auto test/unit/test_winograd.py test/unit/test_linalg.py --durations=20
     - name: Test METAL=1 RANGEIFY=1
       run: METAL=1 RANGEIFY=1 python -m pytest -n=auto test/test_ops.py --durations=20
     - name: Run process replay tests
diff --git a/test/test_pickle.py b/test/test_pickle.py
index d5c62cd77f..ddb01670fd 100644
--- a/test/test_pickle.py
+++ b/test/test_pickle.py
@@ -2,7 +2,7 @@ import unittest, pickle, types
 import numpy as np
 from tinygrad import Tensor, TinyJit, Variable, dtypes
 from tinygrad.helpers import GlobalCounters, ContextVar, Context
-from tinygrad.uop.ops import PatternMatcher, UPat, UOp, Ops
+from tinygrad.uop.ops import PatternMatcher, UPat, UOp
 
 class TestPickle(unittest.TestCase):
   def test_pickle_code_object(self):
@@ -45,10 +45,9 @@ class TestPickle(unittest.TestCase):
     t_values = t.numpy()
     del t # free buffers
     print("** post pickle")
-    init = GlobalCounters.kernel_count
     t2:Tensor = pickle.loads(st)
+    assert t2.uop.is_realized
     np.testing.assert_equal(t_values, t2.numpy())
-    self.assertEqual(GlobalCounters.kernel_count-init, 0)
 
   def test_pickle_realized_tensor_alt2(self):
     print("** init")
@@ -70,14 +69,14 @@ class TestPickle(unittest.TestCase):
   def test_pickle_buffer_uop(self):
     t = Tensor.arange(4).realize()
     a = t.uop
-    assert a.op is Ops.BUFFER
-    self.assertIsNotNone(buffer:=a.realized)
+    assert a.is_realized
+    self.assertIsNotNone(buffer:=a.base.realized)
     s = pickle.dumps(a)
     # free buffers
     del a
     del buffer
     a2:UOp = pickle.loads(s)
-    self.assertListEqual(a2.realized.as_buffer().cast("I").tolist(), [0, 1, 2, 3])
+    self.assertListEqual(a2.base.realized.as_buffer().cast("I").tolist(), [0, 1, 2, 3])
 
   def test_pickle_unrealized_tensor(self):
     t = Tensor.ones(10, 10)
diff --git a/tinygrad/schedule/rangeify.py b/tinygrad/schedule/rangeify.py
index dae1bfc20a..b0405f96de 100644
--- a/tinygrad/schedule/rangeify.py
+++ b/tinygrad/schedule/rangeify.py
@@ -312,7 +312,8 @@ def might_end_axis(idx:UOp):
   if all(x.op not in {Ops.REDUCE_AXIS} for x in idx.toposort()): return None
   to_end_axis = []
   for i,a in enumerate(idx.src[1:]):
-    if any(x.arg > idx.arg for x in a.toposort() if x.op is Ops.RANGE):
+    # in RANGEIFY=1, always realize
+    if not (RANGEIFY > 1) or any(x.arg > idx.arg for x in a.toposort() if x.op is Ops.RANGE):
       to_end_axis.append(i)
   if to_end_axis: return idx.replace(src=(idx.src[0].realize(arg=tuple(to_end_axis)),)+idx.src[1:], arg=None)
   return idx.replace(arg=None)
diff --git a/tinygrad/tensor.py b/tinygrad/tensor.py
index 7269c849f8..615dea1839 100644
--- a/tinygrad/tensor.py
+++ b/tinygrad/tensor.py
@@ -4100,7 +4100,7 @@ class Tensor(MathTrait):
     R = self.clone()
     Q = Tensor.eye(m, dtype=self.dtype).reshape((1,) * len(b_shape) + (m, m)).expand(b_shape + (m, m)).contiguous()
     for i in range(min(m, n)):
-      x = R[..., i:m, i]
+      x = R[..., i:m, i].contiguous()  # TODO: without contigous this can silently be wrong, should at least assert
       s = -x[..., 0].sign()
       u1 = x[..., 0] - s * x.square().sum(-1).sqrt()
       w = x.unsqueeze(-1) / u1.reshape(b_shape + (1, 1))