BackdoorUnlocking NVIDIA Tensor Computations
Chinese developers have found a way to remove the software-imposed limit on tensor computation speed in NVIDIA RTX series graphics cards. This solution allows for nearly double the speed in neural network processing without any hardware modifications.
Test results show a significant performance boost across various GPU models.
- RTX 3060 — up to 99% increase
- RTX 3070 — up to 81% increase
- RTX 3080 — up to 73% increase
- RTX 3090 Ti — up to 80% increase
- RTX 4070 Ti — up to 84% increase
- RTX 4090 — almost double the speed
The limitation was implemented by NVIDIA itself at the software level. Developers believe this is done to stimulate sales of more expensive professional graphics cards designed for artificial intelligence tasks.
It is important to note that unlocking low-level limits may lead to system instability. Ready-made solutions are available via the link below.
18.7K views
