Skip to content

The most powerful and modular diffusion model GUI, api and backend with a graph/nodes interface. The fastest local inference engine in the world.

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

comfyanonymous09dc24c8a982
diff --git a/comfy/model_management.py b/comfy/model_management.py
index 4d13c52c1..7a30c4bec 100644
--- a/comfy/model_management.py
+++ b/comfy/model_management.py
@@ -1087,7 +1087,7 @@ PINNED_MEMORY = {}
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: