I updated the wan2gp launcher to install the new wgp kernels and updated the llama.cpp kernels to version 1.0.29 like announced in Deep Beep Meep's latest Update News.
https://github.com/deepbeepmeep/WanGP#27th-of-october-2026-wangp-v1884---the-one-world-generator
I did a comparison with and without the new wgp kernels.
On an RTX 5080 using sage2 and the wgp kernels with the default workflow and the LTX 2.5 distilled model the total generation time dropped from 30.8 sec to 26.8 sec. That's a speed boost of around 13%. The model was preloaded to get the actual generation time.
26.8 seconds for a 10 sec video is just crazy!
Both wan2gp launcher are updated using the new kernels. To get them just update wan2gp.
In Configuration - Performance in the Int8 Math Kernels setting the wgp kernels should now be displayed as installed.
There was a bug in the GPU detection on Linux which was also fixed. So it should work on all OS now.
I updated the wan2gp launcher to install the new wgp kernels and updated the llama.cpp kernels to version 1.0.29 like announced in Deep Beep Meep's latest Update News.
https://github.com/deepbeepmeep/WanGP#27th-of-october-2026-wangp-v1884---the-one-world-generator
I did a comparison with and without the new wgp kernels.
On an RTX 5080 using sage2 and the wgp kernels with the default workflow and the LTX 2.5 distilled model the total generation time dropped from 30.8 sec to 26.8 sec. That's a speed boost of around 13%. The model was preloaded to get the actual generation time.
26.8 seconds for a 10 sec video is just crazy!
Both wan2gp launcher are updated using the new kernels. To get them just update wan2gp.
In Configuration - Performance in the Int8 Math Kernels setting the wgp kernels should now be displayed as installed.
There was a bug in the GPU detection on Linux which was also fixed. So it should work on all OS now.