Performance regression on Intel Arc B580
- Dominant language
- Python
- Stars
- 133k
- Forks
- 15.7k
- Avg merge
- 1d 7h
- Merged PRs (30d)
- 158
Description
The commit 5aa5ccc9e02aec94cf43e0f71d4b2f62b204b5b6 introduces a performance regression on my B580. The parent commit is fine.
4d6a058bf1dd18fb6d4594081c3f9a7575c97256 performance:
v1-5-pruned-emaonly.safetensors (Load Default)
SDXL (https://comfyanonymous.github.io/ComfyUI_examples/sdxl/sdxl_simple_example.png)
SD3 (https://comfyanonymous.github.io/ComfyUI_examples/sd3/sd3.5_simple_example.png) (`--low-vram`)
5aa5ccc9e02aec94cf43e0f71d4b2f62b204b5b6 performance:
v1-5-pruned-emaonly.safetensors (Load Default)
SDXL (https://comfyanonymous.github.io/ComfyUI_examples/sdxl/sdxl_simple_example.png) (had to use `--low-vram` because it was crashing on the 2~3rd prompt)
SD3 (https://comfyanonymous.github.io/ComfyUI_examples/sd3/sd3.5_simple_example.png) (`--low-vram`. It reloaded the mode, then took a long time to run the prompt then crashed to UR_RESULT_ERROR_OUT_OF_DEVICE_MEMORY)
Containerfile:
```containerfile
ARG ONEAPI_VERSION=2025.3.2-0-devel-ubuntu24.04
FROM docker.io/intel/deep-learning-essentials:$ONEAPI_VERSION
RUN apt update \
&& add-apt-repository --yes ppa:deadsnakes/ppa \
&& apt install --yes python3.11 \
&& ln --symbolic --no-target-directory /usr/bin/python3.11 /usr/bin/python
RUN cd /tmp/ \
&& wget https://bootstrap.pypa.io/get-pip.py \
&& python get-pip.py
ARG COMFYUI_COMMIT
RUN git clone https://github.com/comfyanonymous/ComfyUI \
&& cd /ComfyUI/ && git checkout $COMFYUI_COMMIT
WORKDIR ComfyUI
RUN pip install --pre torch torchvision torchaudio --index-url=https://download.pytorch.org/whl/nightly/xpu \
&& pip install --requirement requirements.txt
#RUN cd /ComfyUI/custom_nodes/ \
# && git clone https://github.com/pamparamm/sd-perturbed-attention \
# && git clone https://github.com/city96/ComfyUI-GGUF && pip install --requirement ComfyUI-GGUF/requirements.txt \
# && git clone https://github.com/Fannovel16/comfyui_controlnet_aux && pip install --requirement comfyui_controlnet_aux/requirements.txt
RUN rm --recursive --force /tmp/* /var/tmp/* \
&& find /var/cache/apt/archives /var/lib/apt/lists -not -name lock -type f -delete \
&& find /var/cache -type f -delete
ENTRYPOINT ["python", "/ComfyUI/main.py"]
```
docker-compose.yaml
```yaml
services:
comfyui:
image: "localhost/comfyui"
command: "--listen 0.0.0.0 --port 8188"
ports:
- "8188:8188"
volumes:
- "./data/models/:/ComfyUI/models/"
- "./data/output/:/ComfyUI/output/"
- "./data/input/:/ComfyUI/input/"
devices:
- "/dev/dri/renderD128"
```
I noticed the newer version takes more time on the "VAE Decode" node
Contributor guide
Assessment
This issue has not been assessed yet.