ComfyUI 5070Ti

Expected Behavior

5070 Ti CUDA error: no kernel image is available for execution on the device 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.

Actual Behavior

CUDA error: no kernel image is available for execution on the device 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.

Steps to Reproduce

CUDA error: no kernel image is available for execution on the device 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.

Debug Logs

CUDA error: no kernel image is available for execution on the device 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.

Other

No response

Apr 24 '25 11:04 kaworukevin

Hey, I don't know but I had this error a few times using blackwell also, but I believe it always had to do with choosing a wrong weight dtype (like fp32).

Apr 24 '25 22:04 OPeixe

What's your PyTorch version? Blackwell (RTX 50xx) GPUs only work with torch >= 2.7

Apr 25 '25 06:04 woct0rdho

you can try this version PyTorch pip3 install --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/cu128

Apr 25 '25 17:04 AiQFeng

pip3 install --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/cu128

doesn't help

Apr 25 '25 20:04 volnodumcev

@volnodumcev Do you use the portable or the desktop version of ComfyUI? Then you need to run the pip command in the Python environment that ComfyUI uses

Also, the stable version of torch 2.7 is released, so you don't need to install the nightly version

Apr 25 '25 23:04 woct0rdho

you can try this version PyTorch pip3 install --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/cu128

Thank you very much 👍 Fixed.

Device RTX 5060 Ti Kernel : Linux 6.14.4 (local build) Debian Trixie GNU/Linux ComfyUI master git today.

Nvidia Libs using

NVIDIA-Linux-x86_64-575.51.02.run --no-kernel-modules

b:00.0 VGA compatible controller: NVIDIA Corporation Device 2d04 (rev a1)

https://github.com/NVIDIA/open-gpu-kernel-modules

modinfo nvidia
filename:       /lib/modules/6.14.4-infonux/kernel/drivers/video/nvidia.ko.xz
import_ns:      DMA_BUF
alias:          char-major-195-*
version:        575.51.02
supported:      external
license:        Dual MIT/GPL
firmware:       nvidia/575.51.02/gsp_tu10x.bin
firmware:       nvidia/575.51.02/gsp_ga10x.bin
srcversion:     403772A6667B6BEAF6B1CB1
alias:          pci:v000010DEd*sv*sd*bc06sc80i00*
alias:          pci:v000010DEd*sv*sd*bc03sc02i00*
alias:          pci:v000010DEd*sv*sd*bc03sc00i00*
alias:          of:N*T*Cnvidia,tegra234-displayC*
alias:          of:N*T*Cnvidia,tegra234-display
depends:        drm
name:           nvidia
retpoline:      Y
vermagic:       6.14.4-infonux SMP preempt mod_unload modversions 
sig_id:         PKCS#7
signer:         Build time autogenerated kernel key
sig_key:        36:00:4C:A0:4C:68:B6:7E:31:82:A3:76:58:99:F9:96:5C:4A:2E:F1
sig_hashalgo:   sha256
signature:      01:AF:90:D2:C0:5C:F8:D2:86:8E:09:FC:A0:53:06:D3:A6:1F:B5:F3: .......

Apr 26 '25 13:04 zz64

you may find this video useful to sucessfully install Comfyui for RTX 50 series, with 12.8 built in https://youtu.be/7JJFJr8doPw?si=9er5r-bTPaN7oGYI

Alternatively for better control of what you are installing, can try manual install of the Pytorch 2.7 with CUDA 12.8 built https://youtu.be/jz7ueLmrvY4?si=oujGB7QPOmkIKMzI

Apr 27 '25 07:04 pocketpoetry

you may find this video useful to sucessfully install Comfyui for RTX 50 series, with 12.8 built in https://youtu.be/7JJFJr8doPw?si=9er5r-bTPaN7oGYI

Alternatively for better control of what you are installing, can try manual install of the Pytorch 2.7 with CUDA 12.8 built https://youtu.be/jz7ueLmrvY4?si=oujGB7QPOmkIKMzI

Using "Open Source software" on "Closed Source OS" ? why ? Sorry, am not Microzoft FAN. I suggest you post videos under GNU/Linux.

May 22 '25 07:05 zz64