NVIDIA’s latest announcement blesses its video-enhancing RTX Video Super Resolution application together with some extra power for LLaMa 2 and Stable Diffusion models.
First thing first. RTX VSR v1.5 is now equipped with a better algorithm leading to better artifact removal and upscaling quality, the best of all – support for RTX 20 series GPU.
No more being left in the dustbin, now every RTX card can run this feature.
The new version also allows the model to run on the display’s native resolution rather than the video source input, further reducing the cases of mushy and blurry pictures.
It is available for all users in the latest GeForce Game Ready Driver update while the Studio Driver gets it next month.
Need some samples? Catch the video below
TensorRT cranking LLM and Stable Diffusion to a new high
For the tools that have dominated the ranks of generative AI in 2023, the NVIDIA TensorRT-LLM acceleration software stack for Windows which also supports local acceleration with up to 4x performance will be launched on the NVIDIA Developer website soon.
On the other hand, the image-generating algorithm Stable Diffusion will now process things at double the speed thanks to TensorRT accelerating the model through layer fusion, precision calibration, kernel auto-tuning, and other capabilities leading to extreme improvement in inference efficiency and speed.
It is available for download right now for those who want to get a taste of it.
New to the scene? Then try the demo to see the entire setup process from preparing to acceleration.