CUDA error out of memory with RTX 4090

[andyjay220]

After installing the 1.1.0 update. I get the following error for text to image, image to video, and text to video generation (even on lowest resolutions and length. I am running locally and have not enabled any new features. I have reinstalled the app cleaning with full resets of all settings several times. I have an RTX 4090 and did local generation of all types without issue on 1.0.5. Here is the error message:
{
"code": "HTTP_500",
"message": "CUDA error: out of memory\nSearch for cudaErrorMemoryAllocation' in https://docs.nvidia.com/cuda/cuda-runtime-api/group__CUDART__TYPES.html for more information.\nCUDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect.\nFor debugging consider passing CUDA_LAUNCH_BLOCKING=1\nCompile with TORCH_USE_CUDA_DSA` to enable device-side assertions."
}

Fixes shown from another post did not work.

[thejeffmillerexperiment]

I am getting the same error
Windows 11
32GB system RAM
RTX 4090

[kypanz]

Hello guys, that happends for low VRAM, check how much VRAM your graphics card have, i think dont work with less than 12 or 16 of VRAM

[andyjay220]

Hello guys, that happends for low VRAM, check how much VRAM your graphics card have, i think dont work with less than 12 or 16 of VRAM

4090 has 24gb VRAM. I could run 1.0.5 no problem. I only have this issue in 1.1.0.

[kypanz]

Hello guys, that happends for low VRAM, check how much VRAM your graphics card have, i think dont work with less than 12 or 16 of VRAM

4090 has 24gb VRAM. I could run 1.0.5 no problem. I only have this issue in 1.1.0.

I understand, the issue could be still the same, u dont have much memory to run it, check the VRAM usage in your OS when u run the generation, how much VRAM you got free before the crash out of memory ?, also if u still having free VRAM send me a image to trying understand better, because if that the case i gonna be curious about how the models are working internally loading in to the VRAM

[andyjay220]

Hello guys, that happends for low VRAM, check how much VRAM your graphics card have, i think dont work with less than 12 or 16 of VRAM

4090 has 24gb VRAM. I could run 1.0.5 no problem. I only have this issue in 1.1.0.

I understand, the issue could be still the same, u dont have much memory to run it, check the VRAM usage in your OS when u run the generation, how much VRAM you got free before the crash out of memory ?, also if u still having free VRAM send me a image to trying understand better, because if that the case i gonna be curious about how the models are working internally loading in to the VRAM

Just to clarify, it's not crashing. The app just throws up the above error message in a pop up window and stops the generation. It's not using any dedicated VRAM when I watch the OS graph. It will just cap out the 16gb shared system RAM and stop the generation. System RAM usage highest was 76%. Changing the fallback policy in Nvidia's control panel didn't change the behavior.

[isvellopez]

Same issue here, when updating from LTX desktop 1.05 to 1.1, it just start to use the share memory, and gives the error you see below

[isvellopez]

I have test many options in windows 11 helped by chatgpt, gemini, etc, but no way to make it work, instead I have configure LTX 2 Web UI that is for low VRAM like mine which only have 16VRAM, anyway, I spend around 2 hours with chatgpt to make work the next link.

https://github.com/nalexand/LTX-2-OPTIMIZED

And I created a manual and a powershell script (with gpt :-)) to make it work in windows 11, I have attached the manual and the script here, and now I'm able to create videos locally again. (I have not test the script, just asked gpt to created it after we make it worked in my laptop to share it)

LTX-2-OPTIMIZED_Manual_&_Script.zip

Anyway, non-technical people probably wont make this work, so I hope devs will solve the LTX Desktop 1.1 error.

[gmuxphil]

Some issue here. Fresh download today. Nvidia 4080

Generated 1 image, fine.
Try to generate video with lowest settings. it thinks about it for a while and then gives the error.
GPU mem never goes above 1GB
Shared GPU mem is used up to 16GB

{
"code": "HTTP_500",
"message": "CUDA error: out of memory\nSearch for cudaErrorMemoryAllocation' in https://docs.nvidia.com/cuda/cuda-runtime-api/group__CUDART__TYPES.html for more information.\nCUDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect.\nFor debugging consider passing CUDA_LAUNCH_BLOCKING=1\nCompile with TORCH_USE_CUDA_DSA` to enable device-side assertions."
}

[gmuxphil]

UPDATE. Latest release works

[bureyburey]

andyjay220 can you confirm 1.2.5 release works for you?
a fix for #141 was released today that should have addressed this as part of the fix

[isvellopez]

It worked again, but this time it takes too much time to generate video.