thank you its 2x performance in my 3090 , with same quality , installation was smooth
thank you its 2x performance in my 3090gpu , with same quality , installation was smooth because you provide whl file and node folder also thanks for your amazing work looking forward for other variant of this model and ltx 2.3 in this format
Were you using pytorch + cu 130+ before?
I am using a very old version pytorch version: 2.5.1+cu121 before and after, but you are right; I need to upgrade, but I am afraid it can mess up with other libraries.
I am using a very old version pytorch version: 2.5.1+cu121 before and after, but you are right; I need to upgrade, but I am afraid it can mess up with other libraries.
Verify that your PyTorch installation for ComfyUI targets CUDA 30 or newer (cu30+). CUDA 30 added native hardware support for int8 convrot, older CUDA builds rely on software emulation, resulting in noticeably slower execution speeds, not sure if this model has another execution route or whatever but its possible.
thank you its 2x performance in my 3090gpu , with same quality , installation was smooth because you provide whl file and node folder also thanks for your amazing work looking forward for other variant of this model and ltx 2.3 in this format
could you please provide me with your workflow?
For me the w4a8 version are extremly slow. So i switched back to my NVFP4 version for my 4090 card