NVIDIA RTX video card owners have found that their GPUs can be significantly more powerful in AI tasks. Enthus

NVIDIA RTX video card owners have found that their GPUs can be significantly more powerful in AI tasks. Enthusiasts have released an open-source GPU-Unlock tool that removes performance limitations of tensor computing without breaking down and replacing the video card.

The results of the tests are impressive:

- RTX 3060 increase by 7899%. - RTX 3070 by 7281%. - RTX 3080 by 6873%. - RTX 3090 Ti up to 80% in BF16. - RTX 4070 Ti by 8384%. - RTX 4090 almost double the speed increase of individual tensor operations. - RTX 2080 Ti SUPER by 3664%, according to the authors of the publication.

It's not about increasing FPS in games, it's about speeding up certain computations used to work with neural networks, generating images and local language models, and the real growth depends on the specific task and can be much lower.

The interesting thing is that the restrictions were apparently software-based, and the authors suggest that NVIDIA could artificially reduce the capabilities of consumer RTXs to separate them from expensive professional accelerators, but there is no official confirmation.

The projects are already available on GitHub, although compatibility and stability are not yet guaranteed. #game