diff --git a/comfy/model_management.py b/comfy/model_management.py
index 4d13c52c10b3..7a30c4becec3 100644
--- a/comfy/model_management.py
+++ b/comfy/model_management.py
@@ -1087,7 +1087,7 @@ def cast_to_device(tensor, device, dtype, copy=False):
 TOTAL_PINNED_MEMORY = 0
 MAX_PINNED_MEMORY = -1
 if not args.disable_pinned_memory:
-    if is_nvidia():
+    if is_nvidia() or is_amd():
         if WINDOWS:
             MAX_PINNED_MEMORY = get_total_memory(torch.device("cpu")) * 0.45  # Windows limit is apparently 50%
         else: