Enable pinned memory by default on Nvidia. (#10656)

Removed the --fast pinned_memory flag.

You can use --disable-pinned-memory to disable it. Please report if it
causes any issues.

Author: comfyanonymous

comfy/cli_args.py

@@ -145,10 +145,11 @@ class PerformanceFeature(enum.Enum):
     Fp8MatrixMultiplication = "fp8_matrix_mult"
     CublasOps = "cublas_ops"
     AutoTune = "autotune"
-    PinnedMem = "pinned_memory"
 
 parser.add_argument("--fast", nargs="*", type=PerformanceFeature, help="Enable some untested and potentially quality deteriorating optimizations. This is used to test new features so using it might crash your comfyui. --fast with no arguments enables everything. You can pass a list specific optimizations if you only want to enable specific ones. Current valid optimizations: {}".format(" ".join(map(lambda c: c.value, PerformanceFeature))))
 
+parser.add_argument("--disable-pinned-memory", action="store_true", help="Disable pinned memory use.")
+
 parser.add_argument("--mmap-torch-files", action="store_true", help="Use mmap when loading ckpt/pt files.")
 parser.add_argument("--disable-mmap", action="store_true", help="Don't use mmap when loading safetensors.")
 

comfy/model_management.py

@@ -1085,22 +1085,21 @@ def cast_to_device(tensor, device, dtype, copy=False):
 
 PINNED_MEMORY = {}
 TOTAL_PINNED_MEMORY = 0
-if PerformanceFeature.PinnedMem in args.fast:
-    if WINDOWS:
-        MAX_PINNED_MEMORY = get_total_memory(torch.device("cpu")) * 0.45  # Windows limit is apparently 50%
-    else:
-        MAX_PINNED_MEMORY = get_total_memory(torch.device("cpu")) * 0.95
-else:
-    MAX_PINNED_MEMORY = -1
+MAX_PINNED_MEMORY = -1
+if not args.disable_pinned_memory:
+    if is_nvidia():
+        if WINDOWS:
+            MAX_PINNED_MEMORY = get_total_memory(torch.device("cpu")) * 0.45  # Windows limit is apparently 50%
+        else:
+            MAX_PINNED_MEMORY = get_total_memory(torch.device("cpu")) * 0.95
+        logging.info("Enabled pinned memory {}".format(MAX_PINNED_MEMORY // (1024 * 1024)))
+
 
 def pin_memory(tensor):
     global TOTAL_PINNED_MEMORY
     if MAX_PINNED_MEMORY <= 0:
         return False
 
-    if not is_nvidia():
-        return False
-
     if not is_device_cpu(tensor.device):
         return False
 
@@ -1121,9 +1120,6 @@ def unpin_memory(tensor):
     if MAX_PINNED_MEMORY <= 0:
         return False
 
-    if not is_nvidia():
-        return False
-
     if not is_device_cpu(tensor.device):
         return False