Open ericko777 opened 1 year ago
Just realise, even if I have 24GB on my 4090GTX... only 8.5GB max is used. Someone know why?
Started from a clean version of Automatic1111... it wasnt like that.... then, when I copied extension from my other folder, it installed Torch 2.0.1... and other stuff... since, restricted to 8.5GB...hum!
Seem it doesnt crash anymore... even 3X in Hires.fix.... started from a clean install
and it use more the 8.5GB... 13.7GB at 3X upscaling
bad news this morning... still crashing... :-/
cuda: 11.8 cudnn: 8800 driver: 545.84
Cuda version is weird, You can try update your cuda version. Refer to: https://docs.nvidia.com/cuda/cuda-toolkit-release-notes/index.html Driver 545.84 must cooperate with cuda 12.3.x.
@thegreenthumb007 ... I now having cudnn: 8906 ... but cannot manage how to change cuda: 11.8 to 12
Even if I tried to install CUDA 12.3 ([cuda_12.3.0_545.84_windows.exe])(https://developer.download.nvidia.com/compute/cuda/12.3.0/local_installers/cuda_12.3.0_545.84_windows.exe)
but it still 11.8... I didnt restart, could it only be that?
device: NVIDIA GeForce RTX 4090 (1) (compute_37) (8, 9) cuda: 11.8 cudnn: 8906 driver: 545.84
I see some benchmark with "torch: 2.2.0.dev20231025+cu121"..... cuda: 12.1 cudnn: 8902 driver: 525.53 24GB
but I dont figure it out how to install it? Do you know?
finaly... install them on ..\AppData\Local\Programs\Python\Python310\Lib\site-packages and copy torch and torchvision to ..\webui\venv\Lib\site-packages
device: NVIDIA GeForce RTX 4090 (1) (sm_90) (8, 9) cuda: 12.1 cudnn: 8801 driver: 545.84
test time!
Same problem :-/
return torch.group_norm(input, num_groups, weight, bias, eps, torch.backends.cudnn.enabled) torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 680.00 MiB. GPU 0 has a total capacty of 23.99 GiB of which 16.57 GiB is free. Of the allocated memory 4.98 GiB is allocated by PyTorch, and 846.17 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting max_split_size_mb to avoid fragmentation. See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF
it work very well with batch of 8 images... but as soon as I use hires.fix... it cannot use all my 24GB...
INFO:sd_dynamic_prompts.dynamic_prompting:Prompt matrix will create 8 images in a total of 1 batches. 100%|██████████████████████████████████████████████████████████████████████████████████| 20/20 [00:05<00:00, 3.90it/s] Total progress: 100%|██████████████████████████████████████████████████████████████████| 40/40 [19:00<00:00, 28.52s/it] Total progress: 100%|██████████████████████████████████████████████████████████████████| 40/40 [19:00<00:00, 1.55it/s]
Working on this pilot... max Hires.fix X2
Pilote GeForce Game Ready - WHQL Version du pilote : 546.01 - Date de sortie : 2023 octobre 31
You can try two ways. 1. Check the CUDA environment variable
Update the cuDNN https://developer.nvidia.com/rdp/cudnn-archive
2. If still failed. For environment clean, I suggest you uninstall GPU,CUDA driver and delete all files. Then reinstall them, of course you can try reinstall some python package.
By the way, my Rtx4090 works will. you might something not install right.
I solved the problem by setting the system Virtual Memory to 60GB
Is there an existing issue for this?
What happened?
Tried to generate an image TXT2IMG with Hires.fix X2 ....
torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 426.00 MiB (GPU 0; 23.99 GiB total capacity; 4.43 GiB already allocated; 17.81 GiB free; 4.56 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation. See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF
GTX 4090 ... 17.81GB Free... Tried to allocate 426.00 MB ... :-/
set COMMANDLINE_ARGS= --opt-sdp-attention --autolaunch --update-all-extensions --api
Steps to reproduce the problem
What should have happened?
Should work!
Sysinfo
app: stable-diffusion-webui.git updated: 2023-11-03 hash: 4afaaf8a url: https://github.com/AUTOMATIC1111/stable-diffusion-webui.git/tree/master arch: AMD64 cpu: Intel64 Family 6 Model 94 Stepping 3, GenuineIntel system: Windows release: Windows-10-10.0.22621-SP0 python: 3.10.6 2.0.0+cu118 autocast half device: NVIDIA GeForce RTX 4090 (1) (compute_37) (8, 9) cuda: 11.8 cudnn: 8800 driver: 545.84 ram: free:16.18 used:15.78 total:31.96 gpu: free:19.42 used:4.57 total:23.99 gpu-active: current:2.94 peak:4.53 gpu-allocated: current:2.94 peak:4.53 gpu-reserved: current:2.95 peak:4.56 gpu-inactive: current:0.01 peak:0.29 events: retries:2 oom:1 utilization: 0 xformers: 0.0.17 diffusers: 0.18.1 transformers: 4.27.4 active: cuda dtype: torch.float16 vae: torch.float16 unet: torch.float16 Memory optimization:None Cross-attention:sdp
What browsers do you use to access the UI ?
Google Chrome
Console logs
Additional information
No response