torch 2.9 incompatible with Pascal architectures

[Juste-Leo2]

EDIT:
It seems that Torch 2.9+cu128 is causing issues now.
This is different from the initial issue, which was fixed in a previous update.
A PR was created at that time; see the conversation below.

---

Hello,

I recently installed ComfyUI Desktop.
Setup: i7-8750H, GTX 1070, 16 GB RAM.
Versions: PyTorch 2.8.0 + cu129

At startup, I noticed warnings about an incompatibility.
When trying to use Flux Schnell, I ran into a kernel-related error.

I found a workaround: since I already had CUDA Toolkit 12.8 installed, I reinstalled torch, torchvision, and torchaudio with version 2.8.0+cu128.
After that, ComfyUI worked successfully.

I’m opening this issue to report this potential problem, and to help beginners or users with older hardware configurations.

Thanks to the ComfyUI team for your great work!

[Juste-Leo2]

same problem here, see #9705 , #9650

[officuse450-cyber]

Hello,

I recently installed ComfyUI Desktop. Setup: i7-8750H, GTX 1070, 16 GB RAM. Versions: PyTorch 2.8.0 + cu129

At startup, I noticed warnings about an incompatibility. When trying to use Flux Schnell, I ran into a kernel-related error.

I found a workaround: since I already had CUDA Toolkit 12.8 installed, I reinstalled torch, torchvision, and torchaudio with version 2.8.0+cu128. After that, ComfyUI worked successfully.

I’m opening this issue to report this potential problem, and to help beginners or users with older hardware configurations.

Thanks to the ComfyUI team for your great work!

Could you give links to torch, torchvision, and torchaudio with version 2.8.0+cu128.

[Juste-Leo2]

Hey! 👋
Here’s the link to install Torch with CUDA 12.8:

pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu128

I closed this issue since the problem has already been fixed.
You can now install ComfyUI Windows Portable from here:
🔗 https://github.com/comfyanonymous/ComfyUI/releases/tag/v0.3.66

Make sure to grab ComfyUI_windows_portable_nvidia_cu128.7z, I recommend installing it directly, especially if you’re using a GTX or a similar GPU

Hope this helps! 😊

[rainthegod]

Hey! 👋 Here’s the link to install Torch with CUDA 12.8:

pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu128
I closed this issue since the problem has already been fixed. You can now install ComfyUI Windows Portable from here: 🔗 https://github.com/comfyanonymous/ComfyUI/releases/tag/v0.3.66

Make sure to grab ComfyUI_windows_portable_nvidia_cu128.7z, I recommend installing it directly, especially if you’re using a GTX or a similar GPU

Hope this helps! 😊

hi. i have gtx 1050ti. have Cuda Errors. i will try now install Cuda 12.6 insted of 13.0

so with this portable version u mentioned it will fix issue? or how you fix this issue bro?

CUDA error: no kernel image is available for execution on the device
Search for cudaErrorNoKernelImageForDevice' in https://docs.nvidia.com/cuda/cuda-runtime-api/group__CUDART__TYPES.html for more information. CUDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect. For debugging consider passing CUDA_LAUNCH_BLOCKING=1 Compile with TORCH_USE_CUDA_DSA` to enable device-side assertions.

[Frediloker]

reinstall the video card driver

[rainthegod]

reinstall the video card driver

thanks bro, will try.
i decided now install Desktop Comfy to avoid python env issues. but if nothing changes will try reinstall drivers.

btw, can you please confirm that even with pascal cards we can use Ai instruments for at least for some low load images generations or upsclales?

[Juste-Leo2]

Hi @rainthegod ,
Yes, you absolutely need to update your graphics drivers for your GTX.
When you run the nvidia-smi command, you need to have at least CUDA 12.8 in order to use it.
For example, with my GTX 1070 I get:

To explain it simply: graphics drivers need to be compatible with libraries like PyTorch. So if your drivers aren’t up to date, PyTorch won’t work.

The CUDA Toolkit is just an extension for developers. I use it personally, but it’s not required at all to use ComfyUI.

I don’t think ComfyUI Desktop fixes this issue (I’m not entirely sure). But I’m certain it will work from the laptop.

And just to confirm: with my 1070 and 16 GB of RAM, you can get Flux Schnell to run, for example (even if it’s a bit limited).

[rainthegod]

Hi @rainthegod , Yes, you absolutely need to update your graphics drivers for your GTX. When you run the nvidia-smi command, you need to have at least CUDA 12.8 in order to use it. For example, with my GTX 1070 I get:

To explain it simply: graphics drivers need to be compatible with libraries like PyTorch. So if your drivers aren’t up to date, PyTorch won’t work.

The CUDA Toolkit is just an extension for developers. I use it personally, but it’s not required at all to use ComfyUI.

I don’t think ComfyUI Desktop fixes this issue (I’m not entirely sure). But I’m certain it will work from the laptop.

And just to confirm: with my 1070 and 16 GB of RAM, you can get Flux Schnell to run, for example (even if it’s a bit limited).

i think my drivers not old because I renew it some time ago but thanks anyway.

Also, its mandatory to install Visual studio with some C++ 2022 and other componets on 10-12gb or not to rub Ai instruments at Comfy?
because i dont have Installed Visual studio.
cant it be issue also?

[Juste-Leo2]

My recommendations @rainthegod :

• Update your graphics drivers: driver
• Use: ComfyUI_windows_portable_12.8

I hope this will help you. 😊

[Juste-Leo2]

Hi @rainthegod , Yes, you absolutely need to update your graphics drivers for your GTX. When you run the nvidia-smi command, you need to have at least CUDA 12.8 in order to use it. For example, with my GTX 1070 I get:

To explain it simply: graphics drivers need to be compatible with libraries like PyTorch. So if your drivers aren’t up to date, PyTorch won’t work.
The CUDA Toolkit is just an extension for developers. I use it personally, but it’s not required at all to use ComfyUI.
I don’t think ComfyUI Desktop fixes this issue (I’m not entirely sure). But I’m certain it will work from the laptop.
And just to confirm: with my 1070 and 16 GB of RAM, you can get Flux Schnell to run, for example (even if it’s a bit limited).

i think my drivers not old because I renew it some time ago but thanks anyway.

Also, its mandatory to install Visual studio with some C++ 2022 and other componets on 10-12gb or not to rub Ai instruments at Comfy? because i dont have Installed Visual studio. cant it be issue also?

As for Visual Studio, it's mainly used to build specific C dependencies.
You won’t need it with the portable version since it’s already included :)

edit: If you plan to do a lot of AI on Windows without ComfyUI, installing Visual Studio can be a good idea.
But it’s mainly used for building Python libraries.
In Python, there are pre-built libraries that avoid all the hassle of compiling on Windows.

[rainthegod]

My recommendations @rainthegod :

• Update your graphics drivers: driver
• Use: ComfyUI_windows_portable_12.8

I hope this will help you. 😊

okay, installing now, but seems i already tried this version. also i need install CUDA exactly 12.8?

look, but how solve issue with PyTorch and Python versions?
that was my main issue at portable version.

because people says here at github for our old Pascals we need old Pytorch like 2.6.0 and Python 3.12.8 like and this helps for them.

tell please what exactly PyTorch and Python you have bro when you running Cmd from older with portable Comfy?

or this was not issue?

[Juste-Leo2]

no, install the latest driver, It is compatible with previous versions of CUDA.

version of my Python dependencies

PS E:\ComfyUI_windows_portable_nvidia_cu128\ComfyUI_windows_portable> python_embeded/python.exe -m pip list
Package                    Version
-------------------------- ------------
aiohappyeyeballs           2.6.1
aiohttp                    3.12.15
aiosignal                  1.4.0
alembic                    1.16.5
annotated-types            0.7.0
attrs                      25.3.0
av                         15.1.0
certifi                    2025.8.3
cffi                       2.0.0
charset-normalizer         3.4.3
colorama                   0.4.6
comfyui-embedded-docs      0.2.6
comfyui_frontend_package   1.27.10
comfyui_workflow_templates 0.1.94
einops                     0.8.1
filelock                   3.19.1
frozenlist                 1.7.0
fsspec                     2025.9.0
greenlet                   3.2.4
huggingface-hub            0.35.3
idna                       3.10
Jinja2                     3.1.6
kornia                     0.8.1
kornia_rs                  0.1.9
Mako                       1.3.10
MarkupSafe                 3.0.3
mpmath                     1.3.0
multidict                  6.6.4
networkx                   3.5
numpy                      2.3.3
packaging                  25.0
pillow                     11.3.0
pip                        25.2
propcache                  0.3.2
psutil                     7.1.0
pycparser                  2.23
pydantic                   2.11.9
pydantic_core              2.33.2
pydantic-settings          2.11.0
pygit2                     1.18.2
python-dotenv              1.1.1
PyYAML                     6.0.3
regex                      2025.9.18
requests                   2.32.5
safetensors                0.6.2
scipy                      1.16.2
sentencepiece              0.2.1
setuptools                 80.9.0
soundfile                  0.13.1
spandrel                   0.4.1
SQLAlchemy                 2.0.43
sympy                      1.14.0
tokenizers                 0.22.1
torch                      2.8.0+cu128
torchaudio                 2.8.0+cu128
torchsde                   0.2.6
torchvision                0.23.0+cu128
tqdm                       4.67.1
trampoline                 0.1.2
transformers               4.56.2
typing_extensions          4.15.0
typing-inspection          0.4.1
urllib3                    2.5.0
yarl                       1.20.1

python version:

PS E:\ComfyUI_windows_portable_nvidia_cu128\ComfyUI_windows_portable> python_embeded/python.exe --version
Python 3.12.10

when I launch run_nvidia_gpu.bat

E:\ComfyUI_windows_portable_nvidia_cu128\ComfyUI_windows_portable>.\python_embeded\python.exe -s ComfyUI\main.py --windows-standalone-build
Checkpoint files will always be loaded safely.
Total VRAM 8192 MB, total RAM 16288 MB
pytorch version: 2.8.0+cu128
Set vram state to: NORMAL_VRAM
Device: cuda:0 NVIDIA GeForce GTX 1070 : cudaMallocAsync
Using pytorch attention
Python version: 3.12.10 (tags/v3.12.10:0cc8128, Apr  8 2025, 12:21:36) [MSC v.1943 64 bit (AMD64)]
ComfyUI version: 0.3.64
ComfyUI frontend version: 1.27.10
[Prompt Server] web root: E:\ComfyUI_windows_portable_nvidia_cu128\ComfyUI_windows_portable\python_embeded\Lib\site-packages\comfyui_frontend_package\static

Import times for custom nodes:
   0.0 seconds: E:\ComfyUI_windows_portable_nvidia_cu128\ComfyUI_windows_portable\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://127.0.0.1:8188

Could you show me your nvidia-smi if it doesn't work after update? Because possibly CUDA 12.8 isn't supported with GTX 1050 Ti

[rainthegod]

thx. look, bro! i think i found also issue with Cuda in my system.

two commands shows different info. but last time i removed 13.0 and installed 12.6 which i have now.

and im installing Nvidia Studio drivers via nvidia app.

what need to do with this Cuda things? i see issue also in this. apps think i use 13.0

maybe just need reboot PC after i intalled 12.6 which i dont do?

[Juste-Leo2]

and im installing Nvidia Studio drivers via nvidia app.

what need to do with this Cuda things? i see issue also in this. apps think i use 13.0

maybe just need reboot PC after i intalled 12.6 which i dont do?

Were you already using the CUDA toolkit?

If that’s the case, you can uninstall CUDA Toolkit 12.6 and install version 12.8 of the CUDA Toolkit.
Otherwise PyTorch will bug, because it will try to interpret everything using CUDA Toolkit 12.6.
These are issues I’ve had in the past.