Pinned mem also seems to work on AMD. (comfyanonymous#10658)

Author: comfyanonymous

comfy/model_management.py

@@ -1120,7 +1120,7 @@ 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: