Published July 22, 2026by Elvis1PR

ComfyUI VRAM OOM on Linux and Windows: LTX 2.3 CUDA Out of Memory Fix?

34 views1 reactions1 comments on CivitAI0 collected

comfyuibug fixtroubleshootingsageattentionpytorchcudamemory managementcomfy aimdortx 4090linuxout of memoryvram leakvramcuda oomcomfyui vram oommemory fixvram oomannouncementcuda out of memoryltx 2.3

Is ComfyUI crashing with a CUDA Out of Memory error on Linux while running LTX 2.3, even though your GPU still has plenty of VRAM?

This article documents a confirmed comfy-aimdo 0.4.9 memory management bug that causes false VRAM exhaustion during LTX 2.3 inference.

Inside you'll learn:

✅ Why the error happens only on certain Linux setups.
✅ Why RTX 4090 24GB can still throw CUDA OOM.
✅ Why updating PyTorch or NVIDIA drivers doesn't solve it.
✅ How nvidia-ml-py can display misleading VRAM usage.
✅ Why --disable-smart-memory is only a temporary workaround.
✅ How installing comfy-aimdo 0.4.10 completely fixes the issue.

Full article: https://short.smartart.live/X3Kj3x

Still have questions? Ask the best ComfyUI forum:

https://short.smartart.live/qk3P9F