lowvram and novram OOM at slightly higher resolutions #5
Description
OS: Arch Linux
Kernel: 6.1.9-arch1-2
GPU: Nvidia GeForce GTX 1060 3GB
Nvidia Driver Version: 525.85.05
CUDA Version: 12.0
When attempting to generate a picture that was 768x768 on novram and lowvram, CUDA experienced an out of memory error. Used "Load Default", with only difference being step count reduced to 8 from 20 for faster testing time. This amount of VRAM should reasonably be able to output higher resolution images (up to around 1152x768~) with low vram optimizations.
executed += recursive_execute(prompt, self.outputs, x, extra_data)
File "/mnt/2TBDa/SDSoftware/comfyfork/ComfyUI/main.py", line 71, in recursive_execute
executed += recursive_execute(prompt, outputs, input_unique_id, extra_data)
File "/mnt/2TBDa/SDSoftware/comfyfork/ComfyUI/main.py", line 76, in recursive_execute
outputs[unique_id] = getattr(obj, obj.FUNCTION)(**input_data_all)
File "/mnt/2TBDa/SDSoftware/comfyfork/ComfyUI/nodes.py", line 101, in decode
return (vae.decode(samples), )
File "/mnt/2TBDa/SDSoftware/comfyfork/ComfyUI/comfy/sd.py", line 311, in decode
pixel_samples = self.first_stage_model.decode(1. / self.scale_factor * samples)
File "/mnt/2TBDa/SDSoftware/comfyfork/ComfyUI/comfy/ldm/models/autoencoder.py", line 94, in decode
dec = self.decoder(z)
File "/home/salt/.local/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1130, in _call_impl
return forward_call(*input, **kwargs)
File "/mnt/2TBDa/SDSoftware/comfyfork/ComfyUI/comfy/ldm/modules/diffusionmodules/model.py", line 637, in forward
h = self.up[i_level].block[i_block](h, temb)
File "/home/salt/.local/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1130, in _call_impl
return forward_call(*input, **kwargs)
File "/mnt/2TBDa/SDSoftware/comfyfork/ComfyUI/comfy/ldm/modules/diffusionmodules/model.py", line 132, in forward
h = nonlinearity(h)
File "/mnt/2TBDa/SDSoftware/comfyfork/ComfyUI/comfy/ldm/modules/diffusionmodules/model.py", line 43, in nonlinearity
return x*torch.sigmoid(x)
RuntimeError: CUDA out of memory. Tried to allocate 576.00 MiB (GPU 0; 2.94 GiB total capacity; 1.45 GiB already allocated; 364.56 MiB free; 2.02 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```