diff --git a/openfold3/core/utils/atomize_utils.py b/openfold3/core/utils/atomize_utils.py
index 79436981..077923ad 100644
--- a/openfold3/core/utils/atomize_utils.py
+++ b/openfold3/core/utils/atomize_utils.py
@@ -762,23 +762,23 @@ def get_token_frame_atoms(
         valid_frame_mask:
             [*, N_token] Mask denoting valid frames
     """
-    # Create pairwise atom mask
-    pair_mask = atom_mask[..., None] * atom_mask[..., None, :]
+    # Pairwise atom mask, kept as bool throughout to avoid materializing
+    # large fp32 [N_atom, N_atom] intermediates.
+    am_bool = atom_mask.bool()
+    pair_mask = am_bool[..., None] & am_bool[..., None, :]
 
-    # Update pairwise atom mask
     # Restrict to atoms within the same chain
     atom_asym_id = broadcast_token_feat_to_atoms(
         token_mask=batch["token_mask"],
         num_atoms_per_token=batch["num_atoms_per_token"],
         token_feat=batch["asym_id"],
     )
-    atom_asym_id_mask = atom_asym_id[..., None] == atom_asym_id[..., None, :]
-    pair_mask = pair_mask * atom_asym_id_mask
+    pair_mask &= atom_asym_id[..., None] == atom_asym_id[..., None, :]
 
-    # Compute distance matrix
+    # Compute distance matrix. Use cdist to avoid materializing N*N*3 intermediate
     # [*, N_atom, N_atom]
-    d = torch.sum(eps + (x[..., None, :] - x[..., None, :, :]) ** 2, dim=-1) ** 0.5
-    d = d * pair_mask + inf * (1 - pair_mask)
+    d = torch.cdist(x, x)
+    d.masked_fill_(~pair_mask, inf)
 
     # Find indices of two closest atoms for start atoms
     # [*, N_token]