diff --git a/tinygrad/nn/optim.py b/tinygrad/nn/optim.py
index 74f3d1297a..aa1d3f40a9 100644
--- a/tinygrad/nn/optim.py
+++ b/tinygrad/nn/optim.py
@@ -34,10 +34,9 @@ class SGD(Optimizer):
   def step(self) -> None:
     for i, t in enumerate(self.params):
       assert t.grad is not None
-      # this is needed since the grads can form a "diamond"
+      # contiguous is needed since the grads can allegedly form a "diamond"
       # TODO: fix this in lazy.py
-      t.grad.realize()
-      g = t.grad + self.wd * t.detach()
+      g = t.grad.contiguous() + self.wd * t.detach()
       if self.momentum:
         self.b[i].assign(self.momentum * self.b[i] + g)  # NOTE: self.b[i] is zero on the first run, no if required
         g = (g + self.momentum * self.b[i]) if self.nesterov else self.b[i]