TensorRT Stable diffusion XL support


Nvidia stated that they worked on an extension for stable diffusion that implements TensorRT efficiently with no resolution limitations, when is that going to release?


TensorRT version: latest
GPU Type: 4070ti
Nvidia Driver Version: latest
CUDA Version: 11.8
CUDNN Version: latest
Operating System + Version: window 11/10
Python Version (if applicable): 3.10.9
TensorFlow Version (if applicable): latest
PyTorch Version (if applicable): 2.0.1+cu118
Baremetal or Container (if container which image + tag):

Steps To Reproduce

Attempted to convert the SDXL model into TensorRT, doesn’t work.

Nvidia stated that they made a TensorRT inference for stable diffusion that has no shape limitations about 3 months ago. So why wouldn’t it work on SDXL? It’s a giant model capable of generating the highest quality of AI images. TensorRT increases performance on older versions of stable diffusion about ~80%<=, which is a dramatic performance increase without any degradation.


This is currently not supported. In future releases, support will be added.
Please stay tuned.

Thank you.