xinyun/ComfyUI
1
0
Fork 0
mirror of https://git.datalinker.icu/comfyanonymous/ComfyUI synced 2026-08-16 01:36:41 +08:00
Code Issues Packages Projects Releases Wiki Activity

Change bf16 check and switch non-blocking to off default with option to force to regain speed on certain classes of iGPUs and refactor xpu check.

Browse Source
This commit is contained in:
Simon Lui 2025-08-11 20:37:25 -07:00
parent 966f3a5206
commit 3a1b18a5be
2 changed files with 14 additions and 7 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
comfy/cli_args.py
View File

@ -132,6 +132,8 @@ parser.add_argument("--reserve-vram", type=float, default=None, help="Set the am
parser.add_argument("--async-offload", action="store_true", help="Use async weight offloading.") parser.add_argument("--async-offload", action="store_true", help="Use async weight offloading.")
parser.add_argument("--force-non-blocking", action="store_true", help="Force ComfyUI to use non-blocking operations for all applicable tensors. This may improve performance on some non-Nvidia systems but can cause issues with some workflows.")
parser.add_argument("--default-hashing-function", type=str, choices=['md5', 'sha1', 'sha256', 'sha512'], default='sha256', help="Allows you to choose the hash function to use for duplicate filename / contents comparison. Default is sha256.") parser.add_argument("--default-hashing-function", type=str, choices=['md5', 'sha1', 'sha256', 'sha512'], default='sha256', help="Allows you to choose the hash function to use for duplicate filename / contents comparison. Default is sha256.")
parser.add_argument("--disable-smart-memory", action="store_true", help="Force ComfyUI to agressively offload to regular ram instead of keeping models in vram when it can.") parser.add_argument("--disable-smart-memory", action="store_true", help="Force ComfyUI to agressively offload to regular ram instead of keeping models in vram when it can.")

19
comfy/model_management.py
View File

@ -78,7 +78,6 @@ try:
torch_version = torch.version.__version__ torch_version = torch.version.__version__
temp = torch_version.split(".") temp = torch_version.split(".")
torch_version_numeric = (int(temp[0]), int(temp[1])) torch_version_numeric = (int(temp[0]), int(temp[1]))
xpu_available = (torch_version_numeric[0] < 2 or (torch_version_numeric[0] == 2 and torch_version_numeric[1] <= 4)) and torch.xpu.is_available()
except: except:
pass pass
@ -102,10 +101,14 @@ if args.directml is not None:
try: try:
import intel_extension_for_pytorch as ipex # noqa: F401 import intel_extension_for_pytorch as ipex # noqa: F401
_ = torch.xpu.device_count()
xpu_available = xpu_available or torch.xpu.is_available()
except: except:
xpu_available = xpu_available or (hasattr(torch, "xpu") and torch.xpu.is_available()) pass
try:
_ = torch.xpu.device_count()
xpu_available = torch.xpu.is_available()
except:
xpu_available = False
try: try:
if torch.backends.mps.is_available(): if torch.backends.mps.is_available():
@ -946,9 +949,11 @@ def pick_weight_dtype(dtype, fallback_dtype, device=None):
return dtype return dtype
def device_supports_non_blocking(device): def device_supports_non_blocking(device):
if args.force_non_blocking:
return True
if is_device_mps(device): if is_device_mps(device):
return False #pytorch bug? mps doesn't support non blocking return False #pytorch bug? mps doesn't support non blocking
if is_intel_xpu(): if is_intel_xpu(): #xpu does support non blocking but it is slower on iGPUs for some reason so disable by default until situation changes
return True return True
if args.deterministic: #TODO: figure out why deterministic breaks non blocking from gpu to cpu (previews) if args.deterministic: #TODO: figure out why deterministic breaks non blocking from gpu to cpu (previews)
return False return False
@ -1282,10 +1287,10 @@ def should_use_bf16(device=None, model_params=0, prioritize_performance=True, ma
return False return False
if is_intel_xpu(): if is_intel_xpu():
if torch_version_numeric < (2, 6): if torch_version_numeric < (2, 3):
return True return True
else: else:
return torch.xpu.get_device_capability(device)['has_bfloat16_conversions'] return torch.xpu.is_bf16_supported()
if is_ascend_npu(): if is_ascend_npu():
return True return True