Skip to content

Re-enable MIOpen (cudnn) for amd cards, default MIOPEN_FIND_MODE=FAST, PYTORCH_MIOPEN_SUGGEST_NHWC=0 #11381

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
alexheretic wants to merge 1 commit into
base: master
Choose a base branch
from
+11 −4
Open

Re-enable MIOpen (cudnn) for amd cards, default MIOPEN_FIND_MODE=FAST, PYTORCH_MIOPEN_SUGGEST_NHWC=0 #11381

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
15 changes: 11 additions & 4 deletions comfy/model_management.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -26,6 +26,7 @@
import platform
import weakref
import gc
import os

class VRAMState(Enum):
DISABLED = 0 #No vram present: no need to move models to vram
Expand Down Expand Up @@ -337,17 +338,23 @@ def amd_min_version(device=None, min_rdna_version=0):
try:
if is_amd():
arch = torch.cuda.get_device_properties(get_torch_device()).gcnArchName
if not (any((a in arch) for a in AMD_RDNA2_AND_OLDER_ARCH)):
torch.backends.cudnn.enabled = False # Seems to improve things a lot on AMD
logging.info("Set: torch.backends.cudnn.enabled = False for better AMD performance.")

try:
rocm_version = tuple(map(int, str(torch.version.hip).split(".")[:2]))
except:
rocm_version = (6, -1)

logging.info("AMD arch: {}".format(arch))
logging.info("ROCm version: {}".format(rocm_version))

if os.getenv('MIOPEN_FIND_MODE') is None:
# MIOpen default search mode can cause significant slowdowns without much benefit
os.environ['MIOPEN_FIND_MODE'] = "FAST"
logging.info("Set: MIOPEN_FIND_MODE=FAST for better AMD performance, change by setting MIOPEN_FIND_MODE.")
if os.getenv('PYTORCH_MIOPEN_SUGGEST_NHWC') is None:
# See https://github.com/ROCm/TheRock/issues/2485#issuecomment-3666986174
os.environ['PYTORCH_MIOPEN_SUGGEST_NHWC'] = "0"
logging.info("Set: PYTORCH_MIOPEN_SUGGEST_NHWC=0 for better AMD performance, change by setting PYTORCH_MIOPEN_SUGGEST_NHWC.")

if args.use_split_cross_attention == False and args.use_quad_cross_attention == False:
if importlib.util.find_spec('triton') is not None: # AMD efficient attention implementation depends on triton. TODO: better way of detecting if it's compiled in or not.
if torch_version_numeric >= (2, 7): # works on 2.6 but doesn't actually seem to improve much
Expand Down