12.10.2024 I can't generate 1536*1536 (or any big Hires) images with an RX580. I'm looking for all kinds of ways AND problems with the second generation (does not clean the memory of the GPU)

[Harbitos]

I need to generate a 1536* 1536 (or any big Hires) image, but it doesn't work. I am ready to do anything for this, even if it is generated for at least an hour. Are there extensions that make this possible?

GPU: RX580 8GB
CPU: intel xeon e3 1270 v3
RAM: 16gb.

Please help me.

And another problem:
I restart the SD every time after the first generation because the video card on the second generation does not clear the memory, after which it writes "Low GPU vram warning" to the console and the memory is clogged more, and generates slower.
Why is this so and can it be fixed? I immediately say that clearing %temp% does not help.

(the first generation is at the bottom, there are no warnings)

venv "D:\Stable Diffusion\stable-diffusion-webui-amdgpu-forge\venv\Scripts\Python.exe"
WARNING: ZLUDA works best with SD.Next. Please consider migrating to SD.Next.
fatal: No names found, cannot describe anything.
Python 3.10.6 (tags/v3.10.6:9c7b4bd, Aug  1 2022, 21:53:49) [MSC v.1932 64 bit (AMD64)]
Version: f2.0.1v1.10.1-1.10.1
Commit hash: 545cb6bf1187a11475ce2b28b3f7f99938cddf3d
Using ZLUDA in D:\Stable Diffusion\stable-diffusion-webui-amdgpu-forge\.zluda
WARNING: no ROCm agent was found!
Launching Web UI with arguments: --use-zluda --no-half --upcast-sampling --precision full --theme dark --skip-version-check --always-batch-cond-uncond --opt-sub-quad-attention --disable-nan-check
You are using PyTorch below version 2.3. Some optimizations will be disabled.
Total VRAM 8192 MB, total RAM 16328 MB
pytorch version: 2.2.1+cu118
Set vram state to: NORMAL_VRAM
Device: cuda:0 AMD Radeon RX 580 2048SP [ZLUDA] : native
VAE dtype preferences: [torch.bfloat16, torch.float32] -> torch.bfloat16
CUDA Using Stream: False
Using pytorch cross attention
Using pytorch attention for VAE
ONNX: version=1.19.2 provider=CPUExecutionProvider, available=['TensorrtExecutionProvider', 'CUDAExecutionProvider', 'CPUExecutionProvider']
ControlNet preprocessor location: D:\Stable Diffusion\stable-diffusion-webui-amdgpu-forge\models\ControlNetPreprocessor
*** Error loading script: pa.py
    Traceback (most recent call last):
      File "D:\Stable Diffusion\stable-diffusion-webui-amdgpu-forge\modules\scripts.py", line 525, in load_scripts
        script_module = script_loading.load_module(scriptfile.path)
      File "D:\Stable Diffusion\stable-diffusion-webui-amdgpu-forge\modules\script_loading.py", line 13, in load_module
        module_spec.loader.exec_module(module)
      File "<frozen importlib._bootstrap_external>", line 883, in exec_module
      File "<frozen importlib._bootstrap>", line 241, in _call_with_frames_removed
      File "D:\Stable Diffusion\stable-diffusion-webui-amdgpu-forge\extensions\sd-webui-prevent-artifact\scripts\pa.py", line 55, in <module>
        sd_hijack_clip.FrozenCLIPEmbedderWithCustomWordsBase.process_tokens = process_tokens
    AttributeError: module 'modules.sd_hijack_clip' has no attribute 'FrozenCLIPEmbedderWithCustomWordsBase'

---
2024-10-12 23:07:54,293 - ControlNet - INFO - ControlNet UI callback registered.
Model selected: {'checkpoint_info': {'filename': 'D:\\Stable Diffusion\\stable-diffusion-webui-amdgpu-forge\\models\\Stable-diffusion\\snowpony_v10.safetensors', 'hash': '7a851477'}, 'additional_modules': ['D:\\Stable Diffusion\\stable-diffusion-webui-amdgpu-forge\\models\\VAE\\sdxl.vae.safetensors'], 'unet_storage_dtype': None}
Using online LoRAs in FP16: False
Running on local URL:  http://127.0.0.1:7860

To create a public link, set `share=True` in `launch()`.
Startup time: 31.6s (prepare environment: 1.9s, import torch: 19.7s, initialize shared: 2.2s, load scripts: 2.7s, initialize google blockly: 0.2s, create ui: 3.0s, gradio launch: 1.7s).
Environment vars changed: {'stream': False, 'inference_memory': 1024.0, 'pin_shared_memory': False}
[GPU Setting] You will use 87.50% GPU memory (7168.00 MB) to load weights, and use 12.50% GPU memory (1024.00 MB) to do matrix computation.
Environment vars changed: {'stream': False, 'inference_memory': 1024.0, 'pin_shared_memory': False}
[GPU Setting] You will use 87.50% GPU memory (7168.00 MB) to load weights, and use 12.50% GPU memory (1024.00 MB) to do matrix computation.
Loading Model: {'checkpoint_info': {'filename': 'D:\\Stable Diffusion\\stable-diffusion-webui-amdgpu-forge\\models\\Stable-diffusion\\snowpony_v10.safetensors', 'hash': '7a851477'}, 'additional_modules': ['D:\\Stable Diffusion\\stable-diffusion-webui-amdgpu-forge\\models\\VAE\\sdxl.vae.safetensors'], 'unet_storage_dtype': None}
[Unload] Trying to free all memory for cuda:0 with 0 models keep loaded ... Done.
StateDict Keys: {'unet': 1680, 'vae': 250, 'text_encoder': 197, 'text_encoder_2': 518, 'ignore': 0}
Working with z of shape (1, 4, 32, 32) = 4096 dimensions.
IntegratedAutoencoderKL Unexpected: ['model_ema.decay', 'model_ema.num_updates']
K-Model Created: {'storage_dtype': torch.float16, 'computation_dtype': torch.float16}
Model loaded in 12.2s (unload existing model: 0.3s, forge model load: 11.9s).
activating extra network lora with arguments [<modules.extra_networks.ExtraNetworkParams object at 0x0000024052A59C30>, <modules.extra_networks.ExtraNetworkParams object at 0x0000024052A5AB00>, <modules.extra_networks.ExtraNetworkParams object at 0x0000024052A59F90>]: AttributeError
Traceback (most recent call last):
  File "D:\Stable Diffusion\stable-diffusion-webui-amdgpu-forge\extensions-builtin\sd_forge_lora\networks.py", line 94, in load_networks
    net = load_network(name, network_on_disk)
  File "D:\Stable Diffusion\stable-diffusion-webui-amdgpu-forge\extensions-builtin\sd_forge_lora\networks.py", line 63, in load_network
    net.mtime = os.path.getmtime(network_on_disk.filename)
AttributeError: 'NoneType' object has no attribute 'filename'

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "D:\Stable Diffusion\stable-diffusion-webui-amdgpu-forge\modules\extra_networks.py", line 135, in activate
    extra_network.activate(p, extra_network_args)
  File "D:\Stable Diffusion\stable-diffusion-webui-amdgpu-forge\extensions-builtin\sd_forge_lora\extra_networks_lora.py", line 45, in activate
    networks.load_networks(names, te_multipliers, unet_multipliers, dyn_dims)
  File "D:\Stable Diffusion\stable-diffusion-webui-amdgpu-forge\extensions-builtin\sd_forge_lora\networks.py", line 96, in load_networks
    errors.display(e, f"loading network {network_on_disk.filename}")
AttributeError: 'NoneType' object has no attribute 'filename'

[Unload] Trying to free 3051.58 MB for cuda:0 with 0 models keep loaded ... Done.
[Memory Management] Target: JointTextEncoder, Free GPU: 7347.49 MB, Model Require: 1559.68 MB, Previously Loaded: 0.00 MB, Inference Require: 1024.00 MB, Remaining: 4763.81 MB, All loaded to GPU.
Moving model(s) has taken 5.07 seconds
[Unload] Trying to free 1024.00 MB for cuda:0 with 1 models keep loaded ... Current free memory is 5587.12 MB ... Done.
[Unload] Trying to free 11609.04 MB for cuda:0 with 0 models keep loaded ... Current free memory is 5586.86 MB ... Unload model JointTextEncoder Done.
[Memory Management] Target: KModel, Free GPU: 7333.84 MB, Model Require: 4897.05 MB, Previously Loaded: 0.00 MB, Inference Require: 1024.00 MB, Remaining: 1412.79 MB, All loaded to GPU.
Moving model(s) has taken 17.84 seconds
  0%|                                                                                           | 0/25 [00:14<?, ?it/s]
Traceback (most recent call last):
  File "D:\Stable Diffusion\stable-diffusion-webui-amdgpu-forge\modules_forge\main_thread.py", line 30, in work
    self.result = self.func(*self.args, **self.kwargs)
  File "D:\Stable Diffusion\stable-diffusion-webui-amdgpu-forge\modules\txt2img.py", line 123, in txt2img_function
    processed = processing.process_images(p)
  File "D:\Stable Diffusion\stable-diffusion-webui-amdgpu-forge\modules\processing.py", line 818, in process_images
    res = process_images_inner(p)
  File "D:\Stable Diffusion\stable-diffusion-webui-amdgpu-forge\modules\processing.py", line 1053, in process_images_inner
    samples_ddim = p.sample(conditioning=p.c, unconditional_conditioning=p.uc, seeds=p.seeds, subseeds=p.subseeds, subseed_strength=p.subseed_strength, prompts=p.prompts)
  File "D:\Stable Diffusion\stable-diffusion-webui-amdgpu-forge\modules\processing.py", line 1430, in sample
    samples = self.sampler.sample(self, x, conditioning, unconditional_conditioning, image_conditioning=self.txt2img_image_conditioning(x))
  File "D:\Stable Diffusion\stable-diffusion-webui-amdgpu-forge\modules\sd_samplers_kdiffusion.py", line 240, in sample
    samples = self.launch_sampling(steps, lambda: self.func(self.model_wrap_cfg, x, extra_args=self.sampler_extra_args, disable=False, callback=self.callback_state, **extra_params_kwargs))
  File "D:\Stable Diffusion\stable-diffusion-webui-amdgpu-forge\modules\sd_samplers_common.py", line 272, in launch_sampling
    return func()
  File "D:\Stable Diffusion\stable-diffusion-webui-amdgpu-forge\modules\sd_samplers_kdiffusion.py", line 240, in <lambda>
    samples = self.launch_sampling(steps, lambda: self.func(self.model_wrap_cfg, x, extra_args=self.sampler_extra_args, disable=False, callback=self.callback_state, **extra_params_kwargs))
  File "D:\Stable Diffusion\stable-diffusion-webui-amdgpu-forge\venv\lib\site-packages\torch\utils\_contextlib.py", line 115, in decorate_context
    return func(*args, **kwargs)
  File "D:\Stable Diffusion\stable-diffusion-webui-amdgpu-forge\k_diffusion\sampling.py", line 595, in sample_dpmpp_2m
    denoised = model(x, sigmas[i] * s_in, **extra_args)
  File "D:\Stable Diffusion\stable-diffusion-webui-amdgpu-forge\venv\lib\site-packages\torch\nn\modules\module.py", line 1511, in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
  File "D:\Stable Diffusion\stable-diffusion-webui-amdgpu-forge\venv\lib\site-packages\torch\nn\modules\module.py", line 1520, in _call_impl
    return forward_call(*args, **kwargs)
  File "D:\Stable Diffusion\stable-diffusion-webui-amdgpu-forge\modules\sd_samplers_cfg_denoiser.py", line 199, in forward
    denoised, cond_pred, uncond_pred = sampling_function(self, denoiser_params=denoiser_params, cond_scale=cond_scale, cond_composition=cond_composition)
  File "D:\Stable Diffusion\stable-diffusion-webui-amdgpu-forge\backend\sampling\sampling_function.py", line 362, in sampling_function
    denoised, cond_pred, uncond_pred = sampling_function_inner(model, x, timestep, uncond, cond, cond_scale, model_options, seed, return_full=True)
  File "D:\Stable Diffusion\stable-diffusion-webui-amdgpu-forge\backend\sampling\sampling_function.py", line 303, in sampling_function_inner
    cond_pred, uncond_pred = calc_cond_uncond_batch(model, cond, uncond_, x, timestep, model_options)
  File "D:\Stable Diffusion\stable-diffusion-webui-amdgpu-forge\backend\sampling\sampling_function.py", line 273, in calc_cond_uncond_batch
    output = model.apply_model(input_x, timestep_, **c).chunk(batch_chunks)
  File "D:\Stable Diffusion\stable-diffusion-webui-amdgpu-forge\backend\modules\k_model.py", line 45, in apply_model
    model_output = self.diffusion_model(xc, t, context=context, control=control, transformer_options=transformer_options, **extra_conds).float()
  File "D:\Stable Diffusion\stable-diffusion-webui-amdgpu-forge\venv\lib\site-packages\torch\nn\modules\module.py", line 1511, in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
  File "D:\Stable Diffusion\stable-diffusion-webui-amdgpu-forge\venv\lib\site-packages\torch\nn\modules\module.py", line 1520, in _call_impl
    return forward_call(*args, **kwargs)
  File "D:\Stable Diffusion\stable-diffusion-webui-amdgpu-forge\backend\nn\unet.py", line 713, in forward
    h = module(h, emb, context, transformer_options)
  File "D:\Stable Diffusion\stable-diffusion-webui-amdgpu-forge\venv\lib\site-packages\torch\nn\modules\module.py", line 1511, in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
  File "D:\Stable Diffusion\stable-diffusion-webui-amdgpu-forge\venv\lib\site-packages\torch\nn\modules\module.py", line 1520, in _call_impl
    return forward_call(*args, **kwargs)
  File "D:\Stable Diffusion\stable-diffusion-webui-amdgpu-forge\backend\nn\unet.py", line 83, in forward
    x = layer(x, context, transformer_options)
  File "D:\Stable Diffusion\stable-diffusion-webui-amdgpu-forge\venv\lib\site-packages\torch\nn\modules\module.py", line 1511, in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
  File "D:\Stable Diffusion\stable-diffusion-webui-amdgpu-forge\venv\lib\site-packages\torch\nn\modules\module.py", line 1520, in _call_impl
    return forward_call(*args, **kwargs)
  File "D:\Stable Diffusion\stable-diffusion-webui-amdgpu-forge\backend\nn\unet.py", line 321, in forward
    x = block(x, context=context[i], transformer_options=transformer_options)
  File "D:\Stable Diffusion\stable-diffusion-webui-amdgpu-forge\venv\lib\site-packages\torch\nn\modules\module.py", line 1511, in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
  File "D:\Stable Diffusion\stable-diffusion-webui-amdgpu-forge\venv\lib\site-packages\torch\nn\modules\module.py", line 1520, in _call_impl
    return forward_call(*args, **kwargs)
  File "D:\Stable Diffusion\stable-diffusion-webui-amdgpu-forge\backend\nn\unet.py", line 181, in forward
    return checkpoint(self._forward, (x, context, transformer_options), None, self.checkpoint)
  File "D:\Stable Diffusion\stable-diffusion-webui-amdgpu-forge\backend\nn\unet.py", line 12, in checkpoint
    return f(*args)
  File "D:\Stable Diffusion\stable-diffusion-webui-amdgpu-forge\backend\nn\unet.py", line 235, in _forward
    n = self.attn1(n, context=context_attn1, value=value_attn1, transformer_options=extra_options)
  File "D:\Stable Diffusion\stable-diffusion-webui-amdgpu-forge\venv\lib\site-packages\torch\nn\modules\module.py", line 1511, in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
  File "D:\Stable Diffusion\stable-diffusion-webui-amdgpu-forge\venv\lib\site-packages\torch\nn\modules\module.py", line 1520, in _call_impl
    return forward_call(*args, **kwargs)
  File "D:\Stable Diffusion\stable-diffusion-webui-amdgpu-forge\backend\nn\unet.py", line 154, in forward
    out = attention_function(q, k, v, self.heads, mask)
  File "D:\Stable Diffusion\stable-diffusion-webui-amdgpu-forge\backend\attention.py", line 335, in attention_pytorch
    out = torch.nn.functional.scaled_dot_product_attention(q, k, v, attn_mask=mask, dropout_p=0.0, is_causal=False)
torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 5.00 GiB. GPU 0 has a total capacity of 8.00 GiB of which 6.57 GiB is free. Of the allocated memory 10.17 GiB is allocated by PyTorch, and 395.65 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)
CUDA out of memory. Tried to allocate 5.00 GiB. GPU 0 has a total capacity of 8.00 GiB of which 6.57 GiB is free. Of the allocated memory 10.17 GiB is allocated by PyTorch, and 395.65 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)

[TheFerumn]

Well. There are few things you can try but don't expect much from your GPU. First of all i don't think using both --upcast-sampling and --precision full commands makes any sense but i might be wrong anyway if you want to generate 1536x1536 on SDXL model it might be hard with your GPU but for the beggining i suggest you to try --always-low-vram command and --all-in-fp16 instead of --precision full and --no-half. With this commands you probably won't be able to use inpainting but it will use less VRAM for model loading. On top of that have your tried Hires. fix option to generate in bigger resolution ? If you will have problems with Hires. fix you can go into settings/upscaling and lower Tile size for upscaler you use.

[Harbitos]

Well. There are few things you can try but don't expect much from your GPU. First of all i don't think using both --upcast-sampling and --precision full commands makes any sense but i might be wrong anyway if you want to generate 1536x1536 on SDXL model it might be hard with your GPU but for the beggining i suggest you to try --always-low-vram command and --all-in-fp16 instead of --precision full and --no-half. With this commands you probably won't be able to use inpainting but it will use less VRAM for model loading. On top of that have your tried Hires. fix option to generate in bigger resolution ? If you will have problems with Hires. fix you can go into settings/upscaling and lower Tile size for upscaler you use.

--always-low-vram command and --all-in-fp16 I set, it didn't help. but it was 3%. By changing the "Tile" settings, the image was generated for me at 100%, but after that the same error occurred. I generated 1024*1024 with Hires. fix x2, I noticed that my generation seemed to be 2 times faster, it usually takes 8 minutes, but it became 3-4 minutes, and without Hires.fix as before - 8 min.

Please take a look at the screenshot, I think that I have configured something wrong, because I have lowered all the parameters that I saw.
Send me a screenshot of these settings.

Another question: what does the Kohya HRFix extension do?

And I also got a blue screen 2 times on which it was written about the lack of memory. (1 time 2 days ago, 1 time today, (when the SD was opened)) (the screen is blue like with an emoji ":(" )

[lshqqytiger]

RX 580 is not good to generate such larger images as it upcasts 16-bit fp into 32-bit in hardware level. That means, it consumes more VRAM than Navi cards. However, you can try img2img upscaling that may use slightly less VRAM.

[TheFerumn]

--always-low-vram command and --all-in-fp16 I set, it didn't help. but it was 3%. By changing the "Tile" settings, the image was generated for me at 100%, but after that the same error occurred. I generated 1024*1024 with Hires. fix x2, I noticed that my generation seemed to be 2 times faster, it usually takes 8 minutes, but it became 3-4 minutes, and without Hires.fix as before - 8 min.

You deleted --precision full and --no-half commands as well ?

Please take a look at the screenshot, I think that I have configured something wrong, because I have lowered all the parameters that I saw. Send me a screenshot of these settings.

I am not sure whats the best. You can lower Tile size even to 16 or 32 and then you can generate few images and check if you have visible seams. If it happens then you have to higher Tile overlap. Just experiment with it a little.

Another question: what does the Kohya HRFix extension do?

Never tried it

And I also got a blue screen 2 times on which it was written about the lack of memory.

That shouldn't happen. Do you have swap memory set on your SSD drive ?

I am pretty sure you will never be able to generate 2048x2048 image on your GPU but there is a little chance for 1536x1536. With Hires fix it shouldn't be a big deal. I was able to generate 1024x1024 images with SDXL model on RX570 4GB VRAM WITHOUT Hires fix and you have 4GB VRAM more than this.

[TheFerumn]

RX 580 is not good to generate such larger images as it upcasts 16-bit fp into 32-bit in hardware level. That means, it consumes more VRAM than Navi cards. However, you can try img2img upscaling that may use slightly less VRAM.

Yes i agree but i noticed using --all-in-fp16 still uses less VRAM on the RX570 than using full precision.

[TheFerumn]

@Harbitos I will suggest you to try to generate in 768x768 and use img2img tab to upscale your image 2x with tweaked Tile size and Tile overlap settings. There is really nothing better you can do. And if you have problems with faces use Adetailer extension but it won't work on --all-in-fp16 with your GPU since it works like inpainting

Edit: Also you can try to go into Settings/VAE and change encode and decode type for TAESD. Its way faster and doesn't change much.
You can also try to uncheck this in Settings/Optimizations if you are willing to sacrifice generation speed but i am not sure how much VRAM it will free for you.

Btw. don't use this command --always-batch-cond-uncond in forge. The setting from screen above is for it so you don't need it in your commandline

[Harbitos]

@Harbitos I will suggest you to try to generate in 768x768 and use img2img tab to upscale your image 2x with tweaked Tile size and Tile overlap settings. There is really nothing better you can do. And if you have problems with faces use Adetailer extension but it won't work on --all-in-fp16 with your GPU since it works like inpainting

Nothing helped me.
1. @lshqqytiger I chose only Resize by in img2img and set 1.5 to be 1536* 1536 - the same error.

2. @TheFerumn yes, I deleted them and replaced them with the ones you said didn't help

3.

I am not sure whats the best. You can lower Tile size even to 16 or 32 and then you can generate few images and check if you have visible seams. If it happens then you have to higher Tile overlap. Just experiment with it a little.

– it didn't help.

4.

That shouldn't happen. Do you have swap memory set on your SSD drive?

– yes, it is set to "automatically". now I have 28GB, and at the time of departure it was 18GB, after closing the SD again 18GB. By the way, today I just generated an image (which my graphics card can generate - 1024*1280 without Hires.fix) and my computer memory froze again, the computer rebooted.

5.

I will suggest you to try to generate in 768x768 and use img2img tab to upscale your image 2x with tweaked Tile size and Tile overlap settings. There is really nothing better you can do. And if you have problems with faces use Adetailer extension but it won't work on --all-in-fp16 with your GPU since it works like inpainting

– I did everything as you said by applying all the settings (generated 768*768 with Hires.fix x2). It generated 100%, but then the same error occurred (the console code is at the bottom).

6. I also asked a question in the issue:
I restart the SD every time after the first generation because the video card on the second generation does not clear the memory, after which it writes "Low GPU vram warning" to the console and the memory is clogged more, and generates slower.
Why is this so and can it be fixed? I immediately say that clearing %temp% does not help.

7. If I can't solve the problem with high resolution, then most of all I want to at least solve problem with "non-cleaning graphics card", it really annoys me.

Moving model(s) has taken 28.95 seconds
100%|██████████████████████████████████████████████████████████████████████████████████| 30/30 [03:49<00:00,  7.66s/it]
[Unload] Trying to free 4540.66 MB for cuda:0 with 0 models keep loaded ... Current free memory is 1884.10 MB ... Done.
[Memory Management] Target: KModel, Free GPU: 1884.10 MB, Model Require: 0.00 MB, Previously Loaded: 4897.05 MB, Inference Require: 1024.00 MB, Remaining: 860.10 MB, All loaded to GPU.
Moving model(s) has taken 0.04 seconds
  3%|██▊                                                                                | 1/30 [01:22<39:45, 82.25s/it]
Traceback (most recent call last):█████████████████████                                | 31/60 [06:06<14:45, 30.52s/it]
  File "D:\Stable Diffusion\stable-diffusion-webui-amdgpu-forge\modules_forge\main_thread.py", line 30, in work
    self.result = self.func(*self.args, **self.kwargs)
  File "D:\Stable Diffusion\stable-diffusion-webui-amdgpu-forge\modules\txt2img.py", line 123, in txt2img_function
    processed = processing.process_images(p)
  File "D:\Stable Diffusion\stable-diffusion-webui-amdgpu-forge\modules\processing.py", line 818, in process_images
    res = process_images_inner(p)
  File "D:\Stable Diffusion\stable-diffusion-webui-amdgpu-forge\modules\processing.py", line 1053, in process_images_inner
    samples_ddim = p.sample(conditioning=p.c, unconditional_conditioning=p.uc, seeds=p.seeds, subseeds=p.subseeds, subseed_strength=p.subseed_strength, prompts=p.prompts)
  File "D:\Stable Diffusion\stable-diffusion-webui-amdgpu-forge\modules\processing.py", line 1446, in sample
    return self.sample_hr_pass(samples, decoded_samples, seeds, subseeds, subseed_strength, prompts)
  File "D:\Stable Diffusion\stable-diffusion-webui-amdgpu-forge\modules\processing.py", line 1550, in sample_hr_pass
    samples = self.sampler.sample_img2img(self, samples, noise, self.hr_c, self.hr_uc, steps=self.hr_second_pass_steps or self.steps, image_conditioning=image_conditioning)
  File "D:\Stable Diffusion\stable-diffusion-webui-amdgpu-forge\modules\sd_samplers_kdiffusion.py", line 190, in sample_img2img
    samples = self.launch_sampling(t_enc + 1, lambda: self.func(self.model_wrap_cfg, xi, extra_args=self.sampler_extra_args, disable=False, callback=self.callback_state, **extra_params_kwargs))
  File "D:\Stable Diffusion\stable-diffusion-webui-amdgpu-forge\modules\sd_samplers_common.py", line 272, in launch_sampling
    return func()
  File "D:\Stable Diffusion\stable-diffusion-webui-amdgpu-forge\modules\sd_samplers_kdiffusion.py", line 190, in <lambda>
    samples = self.launch_sampling(t_enc + 1, lambda: self.func(self.model_wrap_cfg, xi, extra_args=self.sampler_extra_args, disable=False, callback=self.callback_state, **extra_params_kwargs))
  File "D:\Stable Diffusion\stable-diffusion-webui-amdgpu-forge\venv\lib\site-packages\torch\utils\_contextlib.py", line 115, in decorate_context
    return func(*args, **kwargs)
  File "D:\Stable Diffusion\stable-diffusion-webui-amdgpu-forge\k_diffusion\sampling.py", line 595, in sample_dpmpp_2m
    denoised = model(x, sigmas[i] * s_in, **extra_args)
  File "D:\Stable Diffusion\stable-diffusion-webui-amdgpu-forge\venv\lib\site-packages\torch\nn\modules\module.py", line 1511, in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
  File "D:\Stable Diffusion\stable-diffusion-webui-amdgpu-forge\venv\lib\site-packages\torch\nn\modules\module.py", line 1520, in _call_impl
    return forward_call(*args, **kwargs)
  File "D:\Stable Diffusion\stable-diffusion-webui-amdgpu-forge\modules\sd_samplers_cfg_denoiser.py", line 199, in forward
    denoised, cond_pred, uncond_pred = sampling_function(self, denoiser_params=denoiser_params, cond_scale=cond_scale, cond_composition=cond_composition)
  File "D:\Stable Diffusion\stable-diffusion-webui-amdgpu-forge\backend\sampling\sampling_function.py", line 362, in sampling_function
    denoised, cond_pred, uncond_pred = sampling_function_inner(model, x, timestep, uncond, cond, cond_scale, model_options, seed, return_full=True)
  File "D:\Stable Diffusion\stable-diffusion-webui-amdgpu-forge\backend\sampling\sampling_function.py", line 303, in sampling_function_inner
    cond_pred, uncond_pred = calc_cond_uncond_batch(model, cond, uncond_, x, timestep, model_options)
  File "D:\Stable Diffusion\stable-diffusion-webui-amdgpu-forge\backend\sampling\sampling_function.py", line 273, in calc_cond_uncond_batch
    output = model.apply_model(input_x, timestep_, **c).chunk(batch_chunks)
  File "D:\Stable Diffusion\stable-diffusion-webui-amdgpu-forge\backend\modules\k_model.py", line 45, in apply_model
    model_output = self.diffusion_model(xc, t, context=context, control=control, transformer_options=transformer_options, **extra_conds).float()
  File "D:\Stable Diffusion\stable-diffusion-webui-amdgpu-forge\venv\lib\site-packages\torch\nn\modules\module.py", line 1511, in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
  File "D:\Stable Diffusion\stable-diffusion-webui-amdgpu-forge\venv\lib\site-packages\torch\nn\modules\module.py", line 1520, in _call_impl
    return forward_call(*args, **kwargs)
  File "D:\Stable Diffusion\stable-diffusion-webui-amdgpu-forge\backend\nn\unet.py", line 713, in forward
    h = module(h, emb, context, transformer_options)
  File "D:\Stable Diffusion\stable-diffusion-webui-amdgpu-forge\venv\lib\site-packages\torch\nn\modules\module.py", line 1511, in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
  File "D:\Stable Diffusion\stable-diffusion-webui-amdgpu-forge\venv\lib\site-packages\torch\nn\modules\module.py", line 1520, in _call_impl
    return forward_call(*args, **kwargs)
  File "D:\Stable Diffusion\stable-diffusion-webui-amdgpu-forge\backend\nn\unet.py", line 83, in forward
    x = layer(x, context, transformer_options)
  File "D:\Stable Diffusion\stable-diffusion-webui-amdgpu-forge\venv\lib\site-packages\torch\nn\modules\module.py", line 1511, in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
  File "D:\Stable Diffusion\stable-diffusion-webui-amdgpu-forge\venv\lib\site-packages\torch\nn\modules\module.py", line 1520, in _call_impl
    return forward_call(*args, **kwargs)
  File "D:\Stable Diffusion\stable-diffusion-webui-amdgpu-forge\backend\nn\unet.py", line 321, in forward
    x = block(x, context=context[i], transformer_options=transformer_options)
  File "D:\Stable Diffusion\stable-diffusion-webui-amdgpu-forge\venv\lib\site-packages\torch\nn\modules\module.py", line 1511, in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
  File "D:\Stable Diffusion\stable-diffusion-webui-amdgpu-forge\venv\lib\site-packages\torch\nn\modules\module.py", line 1520, in _call_impl
    return forward_call(*args, **kwargs)
  File "D:\Stable Diffusion\stable-diffusion-webui-amdgpu-forge\backend\nn\unet.py", line 181, in forward
    return checkpoint(self._forward, (x, context, transformer_options), None, self.checkpoint)
  File "D:\Stable Diffusion\stable-diffusion-webui-amdgpu-forge\backend\nn\unet.py", line 12, in checkpoint
    return f(*args)
  File "D:\Stable Diffusion\stable-diffusion-webui-amdgpu-forge\backend\nn\unet.py", line 235, in _forward
    n = self.attn1(n, context=context_attn1, value=value_attn1, transformer_options=extra_options)
  File "D:\Stable Diffusion\stable-diffusion-webui-amdgpu-forge\venv\lib\site-packages\torch\nn\modules\module.py", line 1511, in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
  File "D:\Stable Diffusion\stable-diffusion-webui-amdgpu-forge\venv\lib\site-packages\torch\nn\modules\module.py", line 1520, in _call_impl
    return forward_call(*args, **kwargs)
  File "D:\Stable Diffusion\stable-diffusion-webui-amdgpu-forge\backend\nn\unet.py", line 154, in forward
    out = attention_function(q, k, v, self.heads, mask)
  File "D:\Stable Diffusion\stable-diffusion-webui-amdgpu-forge\backend\attention.py", line 335, in attention_pytorch
    out = torch.nn.functional.scaled_dot_product_attention(q, k, v, attn_mask=mask, dropout_p=0.0, is_causal=False)
torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 3.16 GiB. GPU 0 has a total capacity of 8.00 GiB of which 2.58 GiB is free. Of the allocated memory 8.37 GiB is allocated by PyTorch, and 543.29 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)
CUDA out of memory. Tried to allocate 3.16 GiB. GPU 0 has a total capacity of 8.00 GiB of which 2.58 GiB is free. Of the allocated memory 8.37 GiB is allocated by PyTorch, and 543.29 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)

[TheFerumn]

Nothing helped me. 1. @lshqqytiger I chose only Resize by in img2img and set 1.5 to be 1536* 1536 - the same error.

So you can't use Hires fix to 1536 resolution ? What's your settings ? Which upscaler you trying to use and which sampler ?

B. By the way, today I just generated an image (which my graphics card can generate - 1024*1280 without Hires.fix) and my computer memory froze again, the computer rebooted.

Check your GPU temperatures during generation. It totally shouldn't do it... Looks like it might be heating too much or SD agressively allocates memory for some reason. I actually never used Zluda on such old GPU. For RX570 i am only using DirectML and i don't have such issues. You can make experiment with another copy of Forge and check if its better on DirectML but it might be slower.

– I did everything as you said by applying all the settings (generated 768*768 with Hires.fix x2). It generated 100%, but then the same error occurred (the console code is at the bottom).

6. I also asked a question in the issue: I restart the SD every time after the first generation because the video card on the second generation does not clear the memory, after which it writes "Low GPU vram warning" to the console and the memory is clogged more, and generates slower. Why is this so and can it be fixed? I immediately say that clearing %temp% does not help.

7. If I can't solve the problem with high resolution, then most of all I want to at least solve problem with "non-cleaning graphics card", it really annoys me.

Its really old GPU and as i said above. Zluda might not work correctly with such tasks. Also memory leaks are nothing new in SD.