Nvidia RTX 5060 TI compatibility

[OldUncleMike]

Is there an existing issue for this problem?

• I have searched the existing issues

Operating system

Windows

GPU vendor

Nvidia (CUDA)

GPU model

RTX 5060 TI

GPU VRAM

16GB

Version number

5.10.0

Browser

Chrome

Python dependencies

{
"version": "5.10.0",
"dependencies": {
"accelerate" : "1.6.0" ,
"compel" : "2.0.2" ,
"cuda" : "12.4" ,
"diffusers" : "0.33.0" ,
"numpy" : "1.26.3" ,
"opencv" : "4.9.0.80" ,
"onnx" : "1.16.1" ,
"pillow" : "11.0.0" ,
"python" : "3.11.10" ,
"torch" : "2.6.0+cu124" ,
"torchvision" : "0.21.0+cu124",
"transformers": "4.51.3" ,
"xformers" : null
},
"config": {
"schema_version": "4.0.2",
"legacy_models_yaml_path": null,
"host": "127.0.0.1",
"port": 9090,
"allow_origins": [],
"allow_credentials": true,
"allow_methods": [""],
"allow_headers": [""],
"ssl_certfile": null,
"ssl_keyfile": null,
"log_tokenization": false,
"patchmatch": true,
"models_dir": "models",
"convert_cache_dir": "models\.convert_cache",
"download_cache_dir": "models\.download_cache",
"legacy_conf_dir": "configs",
"db_dir": "databases",
"outputs_dir": "outputs",
"custom_nodes_dir": "nodes",
"style_presets_dir": "style_presets",
"workflow_thumbnails_dir": "workflow_thumbnails",
"log_handlers": ["console"],
"log_format": "color",
"log_level": "info",
"log_sql": false,
"log_level_network": "warning",
"use_memory_db": false,
"dev_reload": false,
"profile_graphs": false,
"profile_prefix": null,
"profiles_dir": "profiles",
"max_cache_ram_gb": null,
"max_cache_vram_gb": null,
"log_memory_usage": false,
"device_working_mem_gb": 3,
"enable_partial_loading": true,
"keep_ram_copy_of_weights": true,
"ram": null,
"vram": null,
"lazy_offload": true,
"pytorch_cuda_alloc_conf": "backend:cudaMallocAsync",
"device": "auto",
"precision": "auto",
"sequential_guidance": false,
"attention_type": "auto",
"attention_slice_size": "auto",
"force_tiled_decode": false,
"pil_compress_level": 1,
"max_queue_size": 10000,
"clear_queue_on_startup": false,
"allow_nodes": null,
"deny_nodes": null,
"node_cache_size": 512,
"hashing_algorithm": "blake3_single",
"remote_api_tokens": null,
"scan_models_on_startup": false
},
"set_config_fields": ["legacy_models_yaml_path", "enable_partial_loading", "pytorch_cuda_alloc_conf"]
}

What happened

CUDA sm120 used by RTX 5060TI is not supported by PyTorch. Error:
NVIDIA GeForce RTX 5060 Ti with CUDA capability sm_120 is not compatible with the current PyTorch installation.
The current PyTorch install supports CUDA capabilities sm_50 sm_60 sm_61 sm_70 sm_75 sm_80 sm_86 sm_90.
If you want to use the NVIDIA GeForce RTX 5060 Ti GPU with PyTorch, please check the instructions at https://pytorch.org/get-started/locally/

...

RuntimeError: 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.
...

What you expected to happen

RTX 5060 TI to be supported, soon, please.

How to reproduce the problem

Run the community version with RTX 5060 TI

Additional context

Just purchased and installed new RTX 5060 TI from working RTX 2060 (6GB).
If I can help let me know what to do.
I am on Discord, though I tend not to use it as much as I should
Thanks

Discord username

unclemike2997

[OldUncleMike]

Found a workaround. Thanks to Outremer on Discord, Thanks.
First ,make sure you are using a recent version of Invoke Community Edition, 1.2.1 does not work, 1.5.0 does work.

try this - after opening the invoke app (reset your invoke install) - click on manage, click next, select the latest stable version (V5.10.0 at the moment), click next, select the nvidia 30x and above option, click next, click on the repair mode checkbox and click install.

after this completes, click on the Dev console button in the lower left hand corner of the window and click on "start dev console". run the following command

uv pip install --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/cu128 --upgrade --force-reinstall

When PyTorch becomes stable I'm sure this will be included in the Invoke distribution

[Andyx1976]

When PyTorch becomes stable I'm sure this will be included in the Invoke distribution

I wonder, how many month after everybody else that will be. It's about 3 months now and counting.

a bit further down (it's 3 pages back now and still no rtx50 support in invokeai!) is the post asking for the 5090. Same solution as above. With a bit more explanation, if you need it. Why this is not just simply an option in the launcher by now is completely unclear.

the launcher now (from invoke 5.10) at least installs a current bits and bytes version, simplifying the solution to one step, or two without using the launcher console (which was a good tip btw). (update 5.10.1 breaks everything again, solution doesn't work anymore)

[fan9704]

I use RTX5060TI too I met same problem

[Andyx1976]

Currently working solution for: 5.10.1 It seems using in the console:

uv pip install --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/cu128 --no-deps --force-reinstall
does the trick for 5.10.1.

Howto:

Install the version with the launcher, Apparently (one post below) you have to check "repair" as well during installing. When it's finished click the >_ , bottom left in the launcher, then click "Start dev Console", then paste the command above (exactly) and wait til it's done. It should start downloading about 3GB of stuff)).
Post exactly that, it needs --no-deps or it will crash on launch.

See you in the next update when that doesn't work anymore. Actually, if you have it working you may wanna skip smaller updates, it's not worth the trouble finding a new fix (Every. single. time.)
Thankfully at least the bitsandbytes version the launcher installs is now up to date and fully works out of the (launcher)box with Pytorch128... Baby steps.

[OldUncleMike]

The thing to watch out for is when you want to update Invoke.
For me I had to use the Repair Option (tiny check box, bottom left Invoke Community Edition, before you hit the update button). Then re-apply the fix.
Hope its stable enough to incorporate soon.

[mwenz27]

Found a workaround. Thanks to Outremer on Discord, Thanks. First ,make sure you are using a recent version of Invoke Community Edition, 1.2.1 does not work, 1.5.0 does work.

try this - after opening the invoke app (reset your invoke install) - click on manage, click next, select the latest stable version (V5.10.0 at the moment), click next, select the nvidia 30x and above option, click next, click on the repair mode checkbox and click install.
after this completes, click on the Dev console button in the lower left hand corner of the window and click on "start dev console". run the following command
uv pip install --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/cu128 --upgrade --force-reinstall

When PyTorch becomes stable I'm sure this will be included in the Invoke distribution

I followed the instructions - got a new error code

''module that was compiled using NumPy 1.x cannot be run in NumPy 2.1.2 as it may crash. To support both 1.x and 2.x versions of NumPy, modules must be compiled with NumPy 2.0. Some module may need to rebuild instead e.g. with 'pybind11>=2.12'.''

I am able to get it to work when downgrading numpy with this command

uv pip install numpy<2.0

[Andyx1976]

The thing to watch out for is when you want to update Invoke. For me I had to use the Repair Option (tiny check box, bottom left Invoke Community Edition, before you hit the update button). Then re-apply the fix. Hope its stable enough to incorporate soon.

i have the official response that they will only do that when Pytorch 128 becomes officially stable. Til then it will be interesting because with the 5060 it's not just "a few rich people who can afford a 5090" anymore, that you can just ignore.. You don't get any not RTX50 Nvidia GPU anymore. Btw it maybe worth changing the title of this issue to just RTX50 cards in general, since it works for all of them.
The last issue (we're doing this update&fix game for 2 months now!) is 3 pages back.

[mobius29er]

Currently working solution for: 5.10.1 It seems using in the console:

uv pip install --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/cu128 --no-deps --force-reinstall does the trick for 5.10.1.

Howto:

Install the version with the launcher, Apparently (one post below) you have to check "repair" as well during installing. When it's finished click the >_ , bottom left in the launcher, then click "Start dev Console", then paste the command above (exactly) and wait til it's done. It should start downloading about 3GB of stuff)). Post exactly that, it needs --no-deps or it will crash on launch.

See you in the next update when that doesn't work anymore. Actually, if you have it working you may wanna skip smaller updates, it's not worth the trouble finding a new fix (Every. single. time.) Thankfully at least the bitsandbytes version the launcher installs is now up to date and fully works out of the (launcher)box with Pytorch128... Baby steps.

this works as a workaround thank you.

[fan9704]

Right now pytorch 2.7 cu128 was stable version everyone can try to install it for rtx 50xx

[ecker00]

Related 5090 issue: #7683

[fan9704]

Right now pytorch 2.7 cu128 was stable version everyone can try to install it for rtx 50xx

Try It

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

Ref: Pytorch Get-Started