How to use last nvidia CUDNN v. 9.16 with ComfyUI and get a subtantial performance leap (or potentially break your setup, YMMV) #10758
Description
Hi, I just found out how to restore performance in ComfyUI lost after the CUDNN issue.
I got a performance leap with one of my usual WAN 2.2 I2V workflows from about 45-50 s/it to about 35 s/it.
I used https://developer.download.nvidia.com/compute/cudnn/9.16.0/local_installers/cudnn_9.16.0_windows.exe
If you are on Windows, nvidia GPU (might work in other cases, but that's what I tested on myself) and you are prepared to try bleeding edge stuff with the risk of breaking your setup, look at this post of mine and try to replicate my results:
TLDR: I used last CUDNN 9.16 downloaded from nvidia, I installed it and I symply drag and drop the dlls over the same dlls in torch folder, torch being the yesterday nightly 2.10.0.dev20251114+cu130. I modified a number in the file ops.py, you can see exacly where at this commit: b4f30bd. I put a larger number, something like 91700 in order to skip over the current workaround. Instead my PC exploding or being sucked into a black hole, it worked. :))))))
Cheers!
Edit: most of the information is in the issue I opened at Pytorch repository: pytorch/pytorch#167242
I hope you can retrace my steps successfully! Probably this "will come to a PC near you" with updates in ComfyUI and Pytorch in the next weeks, but if you want to try right now, you can, with the information I provided.