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.

Always enable cuda malloc on cu130 and higher. (#14381)

comfyanonymous5fcf7a4a0f77
diff --git a/cuda_malloc.py b/cuda_malloc.py
index f7651981c..8c4422db8 100644
--- a/cuda_malloc.py
+++ b/cuda_malloc.py
@@ -2,6 +2,7 @@ import os
import importlib.util
from comfy.cli_args import args, PerformanceFeature
import subprocess
+import re
#Can't use pytorch to get the GPU names because the cuda malloc has to be set before the first import.
def get_gpu_names():
@@ -77,11 +78,24 @@ try:
except:
pass
+def get_raw_cuda_version(version_str):
+ match = re.search(r'\+cu(\d+)', version_str)
+ if match:
+ try:
+ return int(match.group(1))
+ except:
+ pass
+ return None
+
if not args.cuda_malloc:
try:
if int(version[0]) >= 2 and "+cu" in version: # enable by default for torch version 2.0 and up only on cuda torch
if PerformanceFeature.AutoTune not in args.fast: # Autotune has issues with cuda malloc
- args.cuda_malloc = cuda_malloc_supported()
+ cuda_version = get_raw_cuda_version(version)
+ if cuda_version is not None and cuda_version >= 130:
+ args.cuda_malloc = True
+ else:
+ args.cuda_malloc = cuda_malloc_supported()
except:
pass