mirror of
https://github.com/Comfy-Org/ComfyUI-Manager.git
synced 2026-11-05 07:47:25 +00:00
[GH-ISSUE #2074] PyTorch 2.8.0+rocm6.4 fails to generate images, but PyTorch 2.7.0+rocm6.3 works correctly #860
Labels
No milestone
No project
No assignees
1 participant
Notifications
Due date
No due date set.
Dependencies
No dependencies set.
Reference
Comfy-Org/ComfyUI-Manager#860
Loading…
Add table
Add a link
Reference in a new issue
No description provided.
Delete branch "%!s()"
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?
Originally created by @markawonge on GitHub (Aug 12, 2025).
Original GitHub issue: https://github.com/Comfy-Org/ComfyUI-Manager/issues/2074
I installed PyTorch using the official installation command from pytorch.org:
pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/rocm6.4When I use the Flux workflow from Workflow Templates in the ComfyUI interface, it fails to generate images properly.
but when I generating images with the default v1.5 checkpoint，it works fine.
Could you please explain why this issue is happening, and is it being caused by PyTorch? Thank you for your response.
`python3 main.py --listen --output-directory /home/kwan/SynologyDrive/output
Adding extra search path checkpoints /home/kwan/stable-diffusion-webui/models/Stable-diffusion
Adding extra search path configs /home/kwan/stable-diffusion-webui/models/Stable-diffusion
Adding extra search path vae /home/kwan/stable-diffusion-webui/models/VAE
Adding extra search path loras /home/kwan/stable-diffusion-webui/models/Lora
Adding extra search path loras /home/kwan/stable-diffusion-webui/models/LyCORIS
Adding extra search path upscale_models /home/kwan/stable-diffusion-webui/models/ESRGAN
Adding extra search path upscale_models /home/kwan/stable-diffusion-webui/models/RealESRGAN
Adding extra search path upscale_models /home/kwan/stable-diffusion-webui/models/SwinIR
Adding extra search path embeddings /home/kwan/stable-diffusion-webui/embeddings
Adding extra search path hypernetworks /home/kwan/stable-diffusion-webui/models/hypernetworks
Adding extra search path controlnet /home/kwan/stable-diffusion-webui/models/ControlNet
Adding extra search path clip /home/kwan/stable-diffusion-webui/models/CLIP
Setting output directory to: /home/kwan/SynologyDrive/output
Checkpoint files will always be loaded safely.
Total VRAM 20464 MB, total RAM 31931 MB
pytorch version: 2.8.0+rocm6.4
AMD arch: gfx1100
ROCm version: (6, 4)
Set vram state to: NORMAL_VRAM
Device: cuda:0 Radeon RX 7900 XT : native
Using pytorch attention
Python version: 3.10.18 (main, Jun 5 2025, 13:14:17) [GCC 11.2.0]
ComfyUI version: 0.3.49
****** User settings have been changed to be stored on the server instead of browser storage. ******
****** For multi-user setups add the --multi-user CLI argument to enable multiple user profiles. ******
ComfyUI frontend version: 1.24.4
[Prompt Server] web root: /home/kwan/miniconda3/envs/cf1/lib/python3.10/site-packages/comfyui_frontend_package/static
Import times for custom nodes:
0.0 seconds: /home/kwan/ComfyUI/custom_nodes/websocket_image_save.py
Context impl SQLiteImpl.
Will assume non-transactional DDL.
No target revision found.
Starting server
To see the GUI go to: http://0.0.0.0:8188
To see the GUI go to: http://[::]:8188
got prompt
model weight dtype torch.float16, manual cast: None
model_type EPS
Using split attention in VAE
Using split attention in VAE
VAE load device: cuda:0, offload device: cpu, dtype: torch.float32
CLIP/text encoder model load device: cuda:0, offload device: cpu, current: cpu, dtype: torch.float16
Requested to load SD1ClipModel
loaded completely 19174.8 235.84423828125 True
Requested to load BaseModel`
@markawonge commented on GitHub (Aug 12, 2025):
I attempted to reinstall Ubuntu 24.04 and ROCm 6.4.3, and then installed PyTorch 2.8.0+rocm6.4 using the official commands from pytorch.org. The previous issue still persists.
@ASmallPotato commented on GitHub (Aug 17, 2025):
Can attest to this,
torch==2.8.0+rocm6.4with flux (tried both gguf and safetensors) generate bad image for me, but SDXL works fine. And can confirm downgrading totorch==2.7.1+rocm6.3can work around this issue.System info
@ltdrdata commented on GitHub (Aug 27, 2025):
This repository only deals with issues related to
ComfyUI-Manager. Since this issue is related to the core, it is not appropriate to post it here.Additionally, it is highly likely that the issue should be resolved on the PyTorch side.