Skip to content

[Issue]: Ram keeps getting filled on AMD Rocm #3995

Open
@CoutHi

Description

@CoutHi

Issue Description

Each generation fills up my ram more and more, vram usage during generation will be around 8-9/16 with my settings but the ram gets slowly filled up after each image, it will start out around ~10G of usage and will climb up to 22-23 where it overflows, and this is purely from generating images, I don't need to load a different model or lora or anything like that. Haven't tested Flux etc. this is my experience with SDXL.

If I for example increase the batch count to 8, all 8 images will be generated without the ram increasing beyond what the initial loading requires, but if I do batches of 1 one after the other each image will just fill the ram more and more without anything getting freed I guess.

Version Platform Description

OS: Fedora 42
Browser: Firefox
GPU: Rx 9070 with Rocm-6.3

17:28:37-695352 INFO Starting SD.Next
17:28:37-697976 INFO Logger: file="/intenso/AI/sdnext/sdnext.log" level=INFO host="fedora-desktop" size=725041 mode=append
17:28:37-698707 INFO Python: version=3.11.0 platform=Linux bin="/intenso/AI/sdnext/venv/bin/python3" venv="/intenso/AI/sdnext/venv"
17:28:37-714981 INFO Version: app=sd.next updated=2025-06-16 hash=72eb0132 branch=master url=https://github.com/vladmandic/sdnext/tree/master ui=main
17:28:38-188912 INFO Version: app=sd.next latest=2025-06-16T19:42:26Z hash=72eb0132 branch=master
17:28:38-191213 INFO Platform: arch=x86_64 cpu= system=Linux release=6.14.11-300.fc42.x86_64 python=3.11.0 locale=('en_IE', 'UTF-8') docker=False
17:28:38-192220 INFO Args: []
17:28:38-282514 INFO ROCm: AMD toolkit detected
17:28:38-284226 WARNING ROCm: no agent was found
17:28:38-284723 INFO ROCm: version=6.3
17:28:38-285190 INFO ROCm: HSA_OVERRIDE_GFX_VERSION auto config skipped: device=None version=None
17:28:38-333758 WARNING Modified files: ['.python-version']
17:28:38-350579 INFO Install: verifying requirements
17:28:38-361262 INFO Extensions: disabled=['Lora']
17:28:38-361881 INFO Extensions: path="extensions-builtin" enabled=['sd-extension-system-info', 'sd-webui-agent-scheduler', 'stable-diffusion-webui-rembg',
'sdnext-modernui', 'sd-extension-chainner']
17:28:38-363175 INFO Extensions: path="extensions" enabled=['sd-webui-prompt-all-in-one']
17:28:38-364139 INFO Startup: standard
17:28:38-364547 INFO Verifying submodules
17:28:42-593549 INFO Extensions enabled: ['sd-extension-system-info', 'sd-webui-agent-scheduler', 'stable-diffusion-webui-rembg', 'sdnext-modernui',
'sd-extension-chainner', 'sd-webui-prompt-all-in-one']
17:28:42-594213 INFO Install: verifying requirements
17:28:42-595688 INFO Installer time: total=5.45 sd-webui-prompt-all-in-one=3.94 latest=0.49 git=0.25 submodules=0.20 base=0.19 torch=0.09 log=0.05 branch=0.05
17:28:42-596381 INFO Command line args: [] args=[]
17:28:50-240766 INFO Torch: torch==2.7.1+rocm6.3 torchvision==0.22.1+rocm6.3
17:28:50-242415 INFO Packages: diffusers==0.34.0.dev0 transformers==4.52.4 accelerate==1.7.0 gradio==3.43.2 pydantic==1.10.21
17:28:58-871866 INFO Device detect: memory=15.9 default=balanced
17:28:58-873797 INFO Engine: backend=Backend.DIFFUSERS compute=rocm device=cuda attention="Split attention" mode=no_grad
17:28:59-363129 INFO Torch parameters: backend=rocm device=cuda config=FP16 dtype=torch.float16 context=no_grad nohalf=False nohalfvae=False upcast=False
deterministic=False tunable=[False, True] fp16=pass bf16=pass optimization="Split attention"
17:28:59-364203 INFO Device: device=AMD Radeon Graphics n=1 hip=6.3.42131-fa1d09cbd
17:28:59-522873 INFO Available VAEs: path="models/VAE" items=0
17:28:59-523692 INFO Available UNets: path="models/UNET" items=0
17:28:59-524273 INFO Available TEs: path="models/Text-encoder" items=0
17:28:59-525179 INFO Available Models: safetensors="models/Stable-diffusion":10 diffusers="models/Diffusers":0 items=10 time=0.00
17:28:59-532904 INFO Available LoRAs: path="models/Lora" items=34 folders=2 time=0.00
17:28:59-547665 INFO Available Styles: path="models/styles" items=288 time=0.01
17:28:59-593856 INFO Available Detailer: path="models/yolo" items=10 downloaded=1
17:28:59-594686 INFO Load extensions
17:29:00-239263 INFO Extension: script='extensions-builtin/sd-webui-agent-scheduler/scripts/task_scheduler.py' Using sqlite file:
extensions-builtin/sd-webui-agent-scheduler/task_scheduler.sqlite3
17:29:00-250342 INFO Extension: script='extensions/sd-webui-prompt-all-in-one/scripts/on_app_started.py' sd-webui-prompt-all-in-one background API service started
successfully.
17:29:00-257014 INFO Available Upscalers: items=71 downloaded=0 user=0 time=0.01 types=['None', 'Resize', 'Latent', 'AsymmetricVAE', 'DCC', 'VIPS', 'ChaiNNer',
'AuraSR', 'SwinIR', 'SCUNet', 'RealESRGAN', 'Diffusion', 'ESRGAN']
17:29:00-260764 INFO UI locale: name="Auto"
17:29:00-262312 INFO UI theme: type=Standard name="invoked" available=13

Relevant log output

Backend

Diffusers

Compute

AMD ROCm

UI

ModernUI

Branch

Master

Model

StableDiffusion XL

Acknowledgements

  • I have read the above and searched for existing issues
  • I confirm that this is classified correctly and its not an extension issue

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions